# K3 control: oba2006 (Oba et al. 2006, Cancer Letters 244:260-267)

Controller: K3 (independent). Full text: `Understudier WCRF/oba2006.txt`. Date: 2026-10-30.
Method: Table 2 meat lines parsed by script and compared field by field with estimates.csv (36 category rows); row 037 and all definitions, models and flags checked by reading.

## Summary
- Rows in estimates.csv: 37. Checked: 37. Numbers confirmed: 37/37 (point, CI, n_cases, tertile medians, p_trend).
- Discrepancies: critical 0, major 0, minor 2 (both in cup_extraction.csv, none in estimates.csv).
- Missing estimates: none.

## Discrepancies

| estimate_id / row | field | in database | in paper (table/page) | severity |
|---|---|---|---|---|
| cup2016_oba2006_colon | cup_contrast | "M 56.6 vs 18.7; W 42.3 vs 10 g" | Women's low-tertile red-meat median is 10.7 g (Table 2, p. 263) | minor (may be CUP's own rounding; the paper's value is 10.7) |
| cup2016_oba2006_colon | estimate_ids_used | empty | CUP's "beef and pork" exposure matches the multivariate red-meat rows: oba2006_016-018 (men), 022-024 (women) | minor (codebook check 6 cannot run while this is empty) |

## Definitions, models and flags (checked by reading)
- Red meat is defined as beef and pork. Processed meat is defined separately as ham, sausage, bacon and yakibuta (p. 261). `red_unprocessed` with includes_processed = no is correct.
- Total meat is not defined. The `unclear` flag is correct.
- Model a ("Age-adjusted RR") maps to m1 and model b (age, height, BMI, pack-years, alcohol, physical activity; Table 2 fn b) maps to m2. is_main_model = yes on the m2 rows is correct.
- Energy: intakes were "logarithmically transformed and adjusted for the total energy" (p. 262; Willett residual method). adj_energy = yes with energy_method = residual is correct for both m1 and m2.
- The adjustment flags for m2 match the footnote. Height is adjusted, but the codebook has no flag for it.
- Row 037 (processed meat, men, excluding the first 3 years: 2.26, 1.26-4.04, P trend <0.01) matches p. 264. The model is not restated in the paper; m2 is an assumption, and the notes say so.
- p_trend "!0.01" is the extracted-text form of "<0.01". It is recorded correctly as <0.01.

## Case-count re-derivation
Each set's three category counts sum to the paper's total: men 111 and women 102 for total meat, red meat and processed meat (35+36+40, 40+39+32, 33+34+44; 41+38+23, 50+25+27, 42+37+23). That gives 213 colon cases, which matches studies.cases_colon. Every reference category is 1.00 with no CI. Every point estimate lies inside its CI.

## studies.csv
Confirmed: n = 30,221 (13,894 men, 16,327 women); follow-up 1 Jan 1993 to 31 Dec 2000 (max 8 y); 169-item FFQ; colon only (rectal cases excluded as too few, p. 261); validation r for red meat 0.54 (men) / 0.21 (women). The Table 2 note gives the women's values in order; red meat is the 9th value, 0.21. outcome_ascertainment = mixed (hospitals cross-checked against the registry) is reasonable.

## Missing estimates
None. The paper has no poultry or fish estimates. The only extra meat estimate (sensitivity analysis, processed meat, men) is row 037. Red and total meat in that analysis are only described as "substantially unchanged", with no numbers.

## Rows confirmed correct
All 37 (oba2006_001-037).

## Internal inconsistencies in the paper
- Methods: tertiles were formed on the distribution "by all subjects" (p. 262), yet Table 2 prints different tertile medians for men and women. Either the tertiles were sex-specific or the medians are within-sex medians of common tertiles. This affects how the doses should be read; the extractor's note already records it.
- The Discussion says red meat "tended to increase the risk" in men (p. 265). Table 2 shows a multivariate RR of 1.03 (0.64-1.66) in the top tertile.

## Verdict
Fit for use. All numbers are correct. Before running check 6, fill in estimate_ids_used for the CUP row.