# K6 control – chao2005 (CPS II Nutrition Cohort, JAMA 2005;293:172–182)

Controller: K6 · 2026-10-30 · Source: `post2015/chao2005.txt` only (no work files or study cards read).

## Method
- Tables 2 (pp. 176–177), 3 (p. 178) and 4 (p. 179) were parsed by script from the full text. Every one of the 177 table-derived rows was compared on point, ci_low, ci_high, n_cases and p_trend (p_trend expected only on the Q5 row).
- Case sums per set were checked against the totals in the text: colon men 665, women 532; proximal 667, distal 408, rectosigmoid + rectum 470.
- Point-in-CI and log-symmetry were checked for every printed CI.
- Doses, labels, denominators, model labels, adjustment sets, main-model flags and pages were checked by script for consistency, then read against the footnotes.
- The 7 text rows were checked by reading pp. 177 and 179.

## Results of the number check
- **177/177 table rows match the paper exactly** on point, CI, n_cases and p_trend. **0 mismatches.**
- **Model 1 and Model 2 rows (60 rows) have no CI.** This is correct: Table 2 prints Model 1/2 RRs without CIs (footnotes † and ‡). Their p-trends (<.001/.04, .09/.40, <.001/.01, .13/.44, .09/.48, .01/.04) are on the Q5 rows as printed.
- **Case counts sum to the totals in every set** (all 32 sets, including the uncoded ratio sets).
- **All printed CIs contain the point estimate.** No log-asymmetry above 15 %.
- **Sex-specific red-meat doses are correct:**
  - dose_point = Table 1 medians (men 100/253/398/612/999; women 43/168/278/416/712 g/wk).
  - n_in_category = Table 1 participants per quintile (men 11,615/13,161/14,697/18,811/11,380; women 11,590/22,804/11,256/21,150/12,146).
  - dose_conversions ÷7 values recomputed correctly (14.3…142.7; 6.1…101.7).
- **The reference category is right everywhere:** Q1 for quintile sets; "low in 1982 and 1992/93" for Table 4.
- **All text rows are found verbatim:** 2.40 (1.30–4.43), P = .002, p. 177; 1.75 (1.04–2.96) and 1.31 (0.79–2.15), p. 179; 1.35 (1.04–1.77), P = .02, p. 177; 1.31 (0.96–1.79), P = .03, p. 177; 1.41 (1.12–1.78) and 1.33 (1.08–1.64), p. 179. Outcomes, sexes and pages are correct. The assumptions (Q5 vs Q1, Model 3, sexes combined for the Comment rows) are clearly flagged in the notes.

## Discrepancies

| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| Q1 rows of chao2005_m3_colon_all_processed, _m3_proximal_colon_all_processed, _m3_distal_colon_all_processed, _m3_rectum_all_processed | dose_low / dose_high | empty | Q1 = "0" for both men and women (Table 2/3 footnote), so 0/0 could be coded as on the sex-specific rows | minor |

No other discrepancies.

## Checked by reading
- **Exposure.** Red meat = `red_total` with includes_processed = yes. This is correct: the definition lists bacon, sausage, hot dogs, ham/bologna/salami/lunchmeat and liver.
  - Processed = `processed`.
  - Poultry and fish = `white_meat`, includes_processed = no (chicken/turkey, fried chicken, fried fish, tuna, other fish).
- **Outcome.** "Rectosigmoid and rectum" (C19.0 + C20.9, n = 470) is coded `rectum`. That is acceptable, and the subgroup tags separate rectosigmoid (214) and rectum proper (246).
- **adjustments.csv.**
  - m1 = age (stratified), total energy (continuous), sex. Correct.
  - m2 adds education, BMI, smoking, physical activity, multivitamins, aspirin (adj_nsaid = yes), beer/wine/liquor (adj_alcohol = yes) and HRT. Correct.
  - m3 = m2 + fruits, vegetables and high-fibre grain foods. adj_fruit_veg = yes and adj_fiber = no is defensible, since these are food variables, not fibre intake.
  - adj_family_history = no is correct: it was "examined and excluded as a potential confounder".
  - energy_method = covariate is correct.
  - m4 (residual-energy sensitivity analysis) is an assumption and is flagged as one. Acceptable.
- **is_main_model.**
  - yes only on Model 3 rows in Tables 2–3. This matches KODEBOK 4.2: Model 3 is the most adjusted model and contains no mediator.
  - The Table 4 long-term rows are `no`. That is a defensible judgment, documented in the notes.
- **studies.csv.** Checked: n = 148,610 (69,664 men + 78,946 women), age 50–74, 68-item FFQ, cases 1667 / 1197 / 470, and sex splits 665+532 and 291+179. All correct.
  - Follow-up mean/median is not printed (followup_type = unclear). Correct.
  - Not checked against CUP: in_cup2016 = no / cup_role = excluded. These come from the assignment list.

## Missing estimates
1. **Ratio of red meat to poultry and fish: 59 estimates, deliberately not coded.**
   - Table 2 (cont): men M1/M2/M3, women M1/M2/M3 and combined M3 = 35.
   - Table 3: proximal, distal and rectum = 15.
   - Table 4: 3 subsites × 3 categories = 9.
   - Abstract: distal 1.53 (1.08–2.18).

   This is a ratio, not a meat amount. KODEBOK now has `dose_unit = ratio` in its extension list, so the database owner should decide whether to add these rows. They are not needed for meat pooling.
2. **Nothing else with numbers.**
   - The beef/pork/lamb main-dish question, doneness, mutual adjustment, and the simultaneous adjustment of poultry and fish for red meat are all "data not shown".

## Confirmed rows
All 184 rows are confirmed on the numbers. All remaining fields are confirmed, except for the 4 minor dose-field items above.

## Internal inconsistencies in the paper (recorded as printed in the DB)
- **Table 2, processed meat, men, Model 3:** Q4 and Q5 are both printed "1.11 (0.80-1.54)". This is probably a duplication typo.
- **Table 3, poultry and fish, distal colon:** Q4 and Q5 are both printed "0.73 (0.53-1.01)". Also probably a duplication typo.
- **Table 2 (cont), ratio, men, Model 2, Q2:** printed "107" (that is, 1.07).
- **Tables 2–4 footnote \*** says "Proportions standardized…" on RR tables. This is boilerplate.
- **The Comment section's 1.41 and 1.33** (age- and energy-adjusted) appear in no table. The sex-specific Model 1 Q5 values are 1.70/1.14 (red) and 1.39/1.28 (processed), so these are presumably unpublished sexes-combined Model 1 estimates.

## Verdict
**Fit for use.** Every number matches. The flags and models are coded consistently with KODEBOK. The only open item is a scope decision: whether to add the ratio exposure.