# K2 control – jarvinen2001 (Järvinen et al., Br J Cancer 2001;85:357–61, Finnish Mobile Clinic)

## Method
- Table 3 (p. 359) parsed by script. I compared all 42 rows (all meat, red meat, poultry, liver × CRC/colon/rectum) on point and CI, and checked every men/women quartile bound in category_label. **0 mismatches.**
- Definitions, model, studies.csv, dose_conversions and cup_extraction were checked by reading.

## Discrepancies

| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| jarvinen2001_025–030 | dose_unit | empty | Poultry is consumers vs non-consumers ("1 (no)", "2 (yes)"). The vocabulary value `ever_never` (used for Tiemersma's sausage) fits. | minor |
| jarvinen2001_025–030 | notes | boilerplate "Sex-specific quartile cut-points …" | Poultry has no quartiles | minor (label) |
| jarvinen2001_001–012 | exposure_includes_processed | `unclear` | Table 1 calls the variable "Meat and meat products"; text p. 358 says "Total consumption of meat and meat products". "Meat products" strongly implies processed products are included. `unclear` is defensible under the KODEBOK rule. `yes` (probable) could be argued, and the note should cite the Table 1 wording. | minor (judgement) |

No critical or major discrepancies.

## Checks that passed (selected)
- **Quartile bounds, red meat vs all meat**, verified against the paper:
  - All meat: men <101 / 101–151 / 152–220 / >220; women <67 / 67–101 / 102–145 / >145 g/d.
  - Red meat: men <94 / 94–141 / 142–206 / >206; women <61 / 61–92 / 93–134 / >134 g/d.
  - The red-meat bounds are 91–94 % of the all-meat bounds (men 0.93/0.93/0.94; women 0.91/0.91/0.92), as the notes and cup_extraction say. Red meat is therefore nearly all of meat, and very probably includes meat products.
  - `red_undefined` / `unclear` is correct, because the paper never defines red meat.
- **Bounds are sex-specific and the RRs are sex-combined,** so dose_low/high are left empty and the bounds go in the label. This is the right solution. The CUP-rule midpoints in dose_conversions are arithmetically correct (e.g. red meat, men: 47 / 117.5 / 174 / 238).
- **Cases:** the paper gives totals only: 109 CRC = 63 colon + 46 rectum (54 men, 55 women). No per-category cases are printed, so n_cases is empty, which is correct. studies.csv has 109/63/46, which is correct. cup_extraction correctly records that CUP used 63 for colorectal (cases_correct 109).
- **Model m1** (Table 3 fn a): age, sex, BMI, occupation, smoking, geographic area, energy, vegetables, fruits, cereals. Flags are correct: fruit_veg = yes, fiber = no, alcohol = no, education = no (occupation was used), region as a covariate. This is the only model for meat, so `is_main_model = yes` is correct.
- **studies.csv:** n 9,959, diet-history interview (>100 items, so diet_items is empty), baseline 1967–72, "up to 32 years" (max), registry, no validation r for meat. All correct.
- **Source page:** Table 3 is on p. 359. Correct.

## Missing estimates
None. Eggs are out of scope. The age- and sex-adjusted results in the text are for protein (0.78 etc.), not meat. No fried-meat or processed-meat estimates are given in this paper (those are Knekt 1994/1999).

## Rows confirmed correct
Every number in all 42 rows is confirmed. Rows 013–024 and 031–042 are confirmed on every field.

## Inconsistencies within the paper
- Red meat CRC Q2 is printed as **1.06 (0.67–2.01)**, which is clearly log-asymmetric. For comparison, all meat Q2 is 1.07 (0.62–1.86). This is probably a misprint. Row 014 flags it correctly.
- Methods p. 357 say the survey ran 1966–1972; the summary says baseline 1967–1972. The diet interviews started in 1967, so the coding of 1967 is fine.