# K10 control – kabat2007 (Kabat et al., Br J Cancer 2007;97:118-22, Canadian NBSS)

Source: historisk/kabat2007.txt (PMC XML-derived, complete). The source_page article range is acceptable.
Rows checked: 55.

## Discrepancies
| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| _031–_045 (alcohol strata) | adjustment_set_id m2: adj_alcohol | yes | Table 3 fn a: "all except the stratification variable". Within alcohol strata, alcohol is the stratifier, not a covariate. | minor |
| _046–_055 (HRT strata) | m2: adj_hrt | yes | HRT is the stratifier here, and the analysis is restricted to postmenopausal women (fn d). m2 is shared by both stratifications, so its flags are right for one and wrong for the other. Split m2 into m2a (alcohol strata) and m2b (HRT strata). | minor |
| _001–_015, _031–_055 | dose_unit | mg_day | not in the dose_unit vocabulary. Heme iron is in mg/day as printed; add the value to NOTES_all. | minor |

No numeric discrepancies. All HRs, CIs and P trends for heme iron and red meat (colon, rectum, colorectal) in Table 2, and for heme iron by alcohol and HRT strata in Table 3, match the text exactly. That includes red-meat rectum Q5 1.95 (1.21–3.16), P 0.008, which the Results prose repeats.

## Re-derivations
- Totals: colon 428 + rectum 195 = 623, which exceeds the 617 colorectal cases because 6 women had both (Table 2 fn b). studies.csv has these right.
- No per-category case counts are printed, so empty n_cases is correct.
- Alcohol strata cases: 142 + 364 + 83 = 589 of 617 (missing alcohol).
- HRT strata cases: 175 + 152 = 327 postmenopausal.

## Exposure coding
- Red meat is not defined in the paper, so red_undefined / unclear is correct.
- Heme iron is included under the codebook's heme_iron rule. Its includes_processed = yes is reasonable, since ham, bacon and luncheon meats feed the heme computation, but note that heme also includes chicken and fish (26%).

## Missing estimates
None within scope. Total iron, meat iron and non-heme iron are nutrients. Proximal and distal results are "data not shown".

## Confirmed
All 55 rows numerically; m1 covariates; studies.csv: 48,666 analysed women aged 40–59, mean follow-up 16.4 years, 786,588 person-years, 617 / 428 / 195 cases.

## Verdict
Fit for use. Rows 016–030 (red meat, quintiles in g/day with cut-points) are usable for dose-response with the CUP open-interval rule.