# K2 control – tiemersma2002 (Tiemersma et al., Cancer Causes Control 2002;13:383–93, MONICA-NL nested case–control)

## Method
- Table 2 (p. 388) parsed by script. I compared all 48 rows (6 exposures × all/women/men × categories) on cases, controls, OR, CI and p-trend. **0 mismatches.**
- Table 5 (p. 391) parsed by script. I compared all 54 joint genotype × exposure rows (point and CI). **0 mismatches.**
- The 5 text and abstract rows (049–053), the definitions, the adjustment set and studies.csv were checked by reading.

## Discrepancies

| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| tiemersma2002_043–048 | exposure_includes_processed | `partial` | Paper: sandwich filling "comprised processed as well as fresh meats" (p. 390). The content is right, but `partial` is not a bool value at estimate level (KODEBOK allows `partial` only in studies.csv). | minor |

No critical or major discrepancies.

## Checks that passed (selected)
- **Exposure:** fresh red meat is "summation of reported beef and pork intake" (p. 385). Processed items are asked separately (meat snacks, sandwiches). So `red_unprocessed` / `no` is correct. Total meat is the sum of all meat-type frequencies, and the paper does not say whether processed items are in it. `unclear` is the right code.
- **Categories and units:** fresh red meat 0–3 / 3.1–4.5 / 5+ times/week (the gap between 4.5 and 5 is printed that way and noted). Total meat 0–3.9 / 4–5.9 / 6+. Poultry and fish 0–1 / 1–4 / 4+ times/month. Sausage yes/no (`ever_never`). Sandwiches 0–1 / 2+ per day. All correct.
- **Case and control counts:** every category set sums to 102 cases / 537 controls (all), 48/245 (women) and 54/292 (men). This matches the analysed 102 cases + 537 controls (p. 386).
- **Model m1:** age, sex, center, total energy, alcohol, body height (Table 2 fn a, Table 5 fn a). Flags are correct. Covariates tested and not retained are listed correctly.
- **is_main_model:** `yes` for Table 2 rows and `no` for Table 5 joint, text subsite, sensitivity and abstract rows. Sandwich is `no` while sausage is `yes`; the choice is documented and defensible.
- **Text rows:**
  - fish HvL colon 0.5 (0.3–0.9), rectum 1.6 (0.7–3.6) and colon women 0.4 (0.1–0.9), p. 386;
  - red meat excluding those under 50 at end of follow-up 2.0 (1.1–3.8), p. 386;
  - abstract poultry+fish in women 0.5 (0.2–1.07), coded `white_meat` with a flag.
  All correct.
- **Table 5:** the database correctly codes the common reference (slow/normal genotype + lowest exposure), so non-reference-genotype rank-1 cells are not 1.0. These rows will fail KODEBOK check 3 by design. That is documented.
- **studies.csv:**
  - design nested_cc, n 639, follow-up 8.5 y (type unclear), age 20–59, enrolment 1987–1991;
  - cases 102/63/39 (rectum = rectum or rectosigmoid);
  - validity r 0.49/0.40 is sex-specific, so validation_r is empty.
  All correct.
- **dose_conversions:** g_day_assigned left empty (no category medians or portion grams). Table 1 control means of 80.4/42.8/25.2/26.9 g/d are correct.
- **cup_extraction:** contrast "≥5 vs 0–3 times/week" and 102 cases agree with the paper.

## Missing estimates
None. Table 3 (smoking) and Table 4 (genotype main effects) contain no meat estimates. The text says adjusting red meat for poultry and fish "did not influence our results" (p. 390), with no numbers. The alternative-reference NAT2 analysis has no numbers either.

## Rows confirmed correct
Every number in all 107 rows is confirmed. All rows except 043–048 are confirmed on every field.

## Inconsistencies within the paper
- **Sex split:** Table 1 gives 45.1 % female among cases (about 46 women) and 46.0 % among controls (about 247). The Table 2 strata sum to 48 women cases / 245 women controls. This is a small inconsistency in the paper.
- **Abstract:** "poultry and fish … women (OR 0.5; 95% CI 0.2–1.07)" matches neither Table 2 women row: poultry 4+ is 0.5 (0.2–1.1) and fish 4+ is 0.5 (0.2–1.0). No combined poultry+fish variable is reported. Row 053 flags this correctly.
- **Cross-study coding note:** "center" (3 towns, as a covariate) is coded `adj_stratified_center = no` here and in Järvinen, but `yes` for Takachi's PHC area covariate. The coding should be harmonised.