# K2 control – takachi2011 (Takachi et al., Asia Pac J Clin Nutr 2011;20:603–12, JPHC)

## Method
- Table 2 (p. 607) and Table 3 (p. 608) parsed by script. All 240 table rows compared:
  - **Table 2:** median, cases, HR, CI and p-trend for 6 exposures × men/women × colon/rectum × Q1–Q5.
  - **Table 3:** cases, HR, CI and p-trend for 6 exposures × men/women × proximal/distal × Q1–Q5.
  - **0 mismatches.**
- I checked by script that Table 3 rows carry the same sex-specific medians as Table 2.
- The 19 text rows (241–259), definitions, adjustment sets, studies.csv and dose_conversions were checked by reading.

## Discrepancies

| estimate_id / table | field | in database | in paper | severity |
|---|---|---|---|---|
| studies.csv takachi2011 | cases_colorectal / cases_colon / cases_rectum | empty | 1,145 / 788 (481 men + 307 women) / 357 (233 + 124), p. 604. The counts are in the notes but not in the fields, so KODEBOK check 5 cannot run and pooling scripts that read case totals get nothing. | major |
| adjustments takachi2011_m1 | adj_energy; notes | `no`; notes say "energy quintiles also entered as covariate" | m1 = age + area only (p. 605). The note is boilerplate copied from m3 and is wrong for m1. Exposure quintiles are of residual energy-adjusted intake, so by the rule used for pietinen1999_m1 (adj_energy = yes via residual), this should be `yes`. m1 has no estimate rows. | minor |
| adjustments takachi2011_m3/m4/m5 | adj_fish | `yes` | Adjustment is for **dried and salted fish** quintiles (an NOC proxy), not total fish. The notes say so. Analyses that use `adj_fish` as "adjusted for fish (negative control/substitution)" will misread it. `unclear`, or a note-driven exception, is safer. | minor |
| adjustments m1–m5 | adj_stratified_center | `yes` | PHC area entered as a covariate. Järvinen and Tiemersma code region/center covariates as `no`. The coding is inconsistent across studies. | minor |
| takachi2011_* | category_set_id format | `<study>_<exposure>_<model>_<outcome>_<sex>` | KODEBOK 3.3 gives `<study>_<model>_<outcome>_<sex>_<exposure>` | minor (format only; keys are unique and consistent within the study) |
| takachi2011_245–248, 251–252 | n_cases | empty | Stratum totals are printed: ≥60/<60 men 259/222, women 152/155 (p. 606); invasive colon men 269, women 186 (p. 607). These are in the notes, not the field. They are not contrast-specific, so leaving the field empty is acceptable. | minor (optional) |

No critical discrepancies.

## Key checks (as asked)

### Red meat definition
- The red meat items are: 3 beef dishes; 6 pork dishes **including pork liver**; **4 processed products (ham, sausage/Weiner, bacon, luncheon meat)**; and chicken liver (p. 604).
- Table 1 lists Processed meat as a sub-row of Red meat.
- So `red_total`, `exposure_includes_processed = yes` on every "Red meat" row is **correct**. The same goes for studies.csv includes_processed = yes.
- The CUP label "red meat (incl. ham, sausage, bacon, luncheon meat, liver)" matches.
- Beef and pork rows are `no`. Pork includes pork liver; that is not processed, so `no` is fine.

### is_main_model
m3 is the Table 2/3 footnote model (adds dried and salted fish). It is `yes` for every exposure × outcome × sex combination. The heme-iron (m4) and SFA (m5) text estimates are `no`, as KODEBOK 4.2 requires (mediators). m2 is only "data not shown". **Correct.**

### Medians per quintile per sex: red meat vs total meat
- **Red meat** (Table 2 = Table 1):
  - men 15 / 31 / 46 / 65 / 102;
  - women 14 / 29 / 43 / 60 / 93 g/d.
  - Database rows 021–040 and 141–160, and dose_conversions dc01–dc08, have exactly these values. **Correct.**
- **Total meat** (Table 2):
  - men 20 / 39 / 56 / 77 / 117;
  - women 18 / 36 / 52 / 70 / 107 g/d.
  - Rows 001–020 and 121–140 have these values. **Correct, and not mixed up with the red-meat values.**
- Precise Q1 medians in the text (p. 605): 15.4 (men) and 13.6 (women). The "corrected" Q5 medians from weighed records (p. 609): 120 (men) and 105 (women). Both are correctly recorded in dose_conversions.
- Table 1 also prints *mean* total meat **by red-meat quintile** (men 20 / 39 / 56 / 77 / **127**; women 17 / 36 / 52 / 71 / **115**). These are not the total-meat quintile medians, and the database correctly does not use them.
- cup_extraction correctly identifies that CUP's men contrast "117 vs 20" is the total-meat medians, not red meat (102 vs 15).

### Quintile bounds and denominators (red meat)
- Bounds: men <24.1 / 24.1–38.8 / 38.8–54.6 / 54.6–78.7 / ≥78.7; women <22.0 / 22.0–35.8 / 35.8–50.4 / 50.4–71.8 / ≥71.8.
- n per quintile: 7,692/7,693 (men) and 8,439/8,440 (women), from Table 1.
- **Correct.**

### Case counts
Every category set sums exactly to the paper's totals:
- colon 481 men / 307 women;
- rectum 233 / 124;
- proximal 200 / 179;
- distal 257 / 110.

Proximal + distal is 24 short of total colon in men and 18 short in women (unspecified subsite). The extraction notes record this.

## Text rows verified (241–259)

All points, CIs and trend p-values agree with the paper:
- heme-adjusted total meat, men 1.30 (0.87, 1.93; 0.38);
- SFA-adjusted total meat, men 1.43 (0.95, 2.16; 0.19);
- red meat, women: heme-adjusted 1.62 (1.01, 2.61; 0.02); SFA-adjusted 1.38 (0.84, 2.27; 0.18) (all p. 605);
- age strata 1.46 / 1.05 / 1.64 / 1.34;
- processed meat in never/ever smokers 1.79 / 1.10 (p. 606);
- invasive colon 1.19 / 1.39 / 1.30 (combined, sex = all) (p. 607);
- decile results 1.37 / 1.67 (processed colon, p. 607); 0.83 / 1.33 (red rectum); 0.68 / 1.28 (processed rectum) (p. 608).

## Missing estimates
None. Other analyses (BMI, alcohol, cohort strata, excluding the first 2 years, HRT/statins, sex-combined quintiles) are "data not shown".

## studies.csv (other fields)
All correct:
- n 80,658 (38,462 men / 42,196 women);
- age 45–74;
- 138-item FFQ;
- baseline 1995–1999;
- person-years 758,116;
- validation r values;
- outcome ascertainment mixed (active notification + registries).

followup_years is empty. The paper gives no mean or median, and the notes give the computed 9.4 y. Acceptable.

## Rows confirmed correct
Every number in all 259 estimate rows is confirmed. Apart from the format issue in category_set_id, all rows are confirmed on every field.

## Inconsistencies within the paper
- The abstract says the FFQ was given "During 1995-1998"; the methods say 1995–1999 (p. 604); the table titles say "1995 and 1998".
- The Discussion (p. 609) speaks of "quartile groups" and the "highest quartiles of processed meat", but the analysis uses quintiles.
- The Discussion gives "corrected" Q5 medians of 120/105 g/d against the table's 102/93. These are different quantities, and both are recorded.
- Proximal + distal < total colon (24 men, 18 women unassigned), with no explanation.