# K2 control – pietinen1999 (Pietinen et al., Cancer Causes Control 1999;10:387–96, ATBC)

## Method
- Table 5 (p. 393) parsed by script. I compared all 40 rows (5 exposures × 2 models × 4 quartiles) on median, cases, RR, CI and p-trend. **0 mismatches.**
- Definitions, models, studies.csv and dose_conversions were checked by reading.

## Discrepancies
No discrepancies in existing rows.

## Missing estimates

| Table / page | Row | What it is |
|---|---|---|
| Table 5, p. 393 | "Ratio of meat/(poultry + fish)", medians 1.3 / 2.3 / 3.6 / 7.4, cases 48/47/46/44; age-adjusted 1.0 / 1.0 (0.7–1.5) / 1.0 (0.7–1.5) / 1.0 (0.7–1.5), p 0.93; multivariate 1.0 / 1.1 (0.7–1.6) / 1.0 (0.7–1.5) / 1.1 (0.7–1.6), p 0.81 | Red meat vs poultry+fish ratio. English 2004's equivalent ratio *is* coded (`contrast_type = substitution`), so leaving this one out is inconsistent across studies. 8 rows (2 reference rows + 6 estimates). |
| Table 5, p. 393 | "Fried meat, frequency per year", medians 60 / 108 / 144 / 204, cases 53/51/36/45; age 1.0 / 1.0 (0.7–1.4) / 0.7 (0.4–1.0) / 0.9 (0.6–1.3), p 0.27; MV 1.0 / 1.0 (0.7–1.4) / 0.7 (0.4–1.0) / 0.9 (0.6–1.3), p 0.25 | Frequency of fried meat dishes (15 foods: steaks, pork chops etc., p. 388). There is no vocabulary value for it; it is a cooking-method exposure. Optional, but useful as a sensitivity exposure. 8 rows. |
| Table 2, p. 390 | Meat protein quartiles (25.3–53.6 g) | A nutrient, not a meat exposure. Probably out of scope; listed for completeness. |

## Checks that passed (selected)
- **Exposure definitions:**
  - "Beef, pork, lamb" → `red_unprocessed`, includes_processed = no. Correct.
  - "Processed meat" (mostly sausages, p. 394) → `processed`. Correct.
  - "Total red meat" = beef, lamb and pork *combined with* processed meat (p. 391) → `red_total`, includes_processed = yes. Correct.
  - studies.csv has includes_processed = yes for "Total red meat", with a note that the unprocessed variable also exists. Correct.
- **Cases:** every category set sums to 185, which matches the 185 analysed CRC cases (114 colon + 86 rectum − 3 carcinoids − 3 anal − 9 without diet data, p. 388). No colon/rectum split after exclusions, so those fields are empty. Correct.
- **Medians:** red_unprocessed 35/52/69/99; processed 26/50/73/122; total red 79/114/143/203; poultry 0/6/13/27; fish 13/26/40/68 g/d. These are medians of energy-adjusted (residual) intakes, and dose_conversions says so. Correct.
- **Models:**
  - m1 = age (5-y categories) + supplement group.
  - m2 = m1 + smoking years, BMI, alcohol, education, occupational physical activity, calcium (Table 5 fn **).
  - Flags are correct. Energy enters through residual-method quartiles, not as a covariate.
  - `is_main_model`: MV = yes, age = no. Correct.
- **cup_extraction:** rows 005–008 (MV beef/pork/lamb) and contrast "99 vs 35 g/day" are correct.
- **studies.csv:** men only, smokers, 50–69 y, n 27,111 (= 185 + 26,926 in Table 1), mean 8 y, 276 items with photo booklet, registry. Correct.
- **Source page:** Table 5 on p. 393. Correct.
- **Asymmetry flags:** rows 002, 006, 016, 019, 022, 023, 034 and 038 are flagged. These are one-decimal rounding cases, not typos.

## Rows confirmed correct
All 40 rows (pietinen1999_001–040) are confirmed on every field.

## Inconsistencies within the paper
- **Text vs table:** the text (p. 391) says multivariate adjustment did not change the results, so "only the base model … is shown". Table 5 nevertheless prints both the age-adjusted and the multivariate rows. Both are in the database, so this does not matter.
- Table 1 gives a median "Red meat (g)" of 125 (cases) and 128 (noncases). This variable does not obviously match either Table 5 red-meat variable (total red meat median Q2–Q3 is about 114–143). It is not an estimate; it is noted only for anyone using Table 1.