# K3 control: bostick1994 (Bostick et al. 1994, Cancer Causes Control 5:38-52)

Controller: K3 (independent). Full text: `Understudier WCRF/bostick1994.txt`. Date: 2026-10-30.
Method: the meat block of Table 3 (8 exposures × age-adjusted and multivariate models, p. 43-44) was parsed by script and compared with all 74 rows (cases, RR, CI, P for trend). The 6 mismatches the script raised were faults in my own parser (it misread the "--" placeholder in 4-category rows). I re-checked those rows by reading, and the database values are correct. Definitions, models and flags were checked by reading.

## Summary
- Rows: 74. Checked: 74. Numbers confirmed: 74/74.
- Discrepancies: critical 0, major 1, minor 2.
- Missing estimates: none.

## Discrepancies

| estimate_id | field | in database | in paper (table/page) | severity |
|---|---|---|---|---|
| bostick1994_013-016 and 021-024 | is_main_model (with exposure = poultry) | Two separate `poultry` category sets, "with skin" and "without skin", both is_main_model = yes for poultry × colon × women | Table 3, p. 43, has two distinct variables and no combined poultry estimate | major (flag). A pool that selects exposure = poultry and is_main_model = yes will take both sets from one cohort. Distinguish them, for example with subgroup = `with_skin` / `without_skin`, or with a pool rule, and document which one is the negative control |
| bostick1994_065-074 | exposure | total_meat | "Total eggs and meat" includes eggs (Table 3 fn g) | minor. The notes already say so. These rows should be excluded from total_meat pools, or given their own vocabulary value |
| studies.csv bostick1994 | followup_years / followup_type | empty / unclear | Follow-up 1986 to 31 Dec 1990, "over a five-year period", 167,447 person-years (p. 41) | minor. Could be 5 / max (mean is about 4.75 = 167,447 / 35,216) |

## What the main-dish item includes
Table 3 footnote g lists the FFQ items. Among them, "beef, pork, or lamb as a main dish (e.g., steak, roast, ham, etc.)". Hamburger, "beef, pork, or lamb as a sandwich or mixed dish", bacon, hot dogs and other processed meats are separate items.

So the main-dish variable explicitly includes ham, and excludes hamburger and mixed dishes. `red_main_dish` with exposure_includes_processed = `partial` is correct, as are studies.red_meat_definition and includes_processed = partial. CUP's label "red meat main dish (incl. ham)" is consistent with this.

The separate "Red meat" variable (mean 7.7-7.9 servings/week, against 2.4 for main dish) has no definition in the paper. `red_undefined` / `unclear` is correct.

## Category contrasts
- **Main dish:** <1, 1, 2-3 and >3 servings/week, with 44, 73, 60 and 35 cases. Recorded correctly:
  - rank 1 dose_high = 1;
  - rank 4 dose_low = 3, open.
  - The contrast is ">3 vs <1". CUP's contrast and issue text describe it correctly.
- **Other variables:**
  - Poultry: 0, 0.5, 1 and >1.
  - Seafoods: <1, 1, 1.5, 2-2.5 and >2.5.
  - Processed meats: 0, 0.5, 1, 2-3 and >3.
  - Red meat: <4.0, 4.0-6.0, 6.5-8.0, 8.5-11.0 and >11.0.
  - White meat: <1.0, 2.0, 2.5-3.5, 4.0-5.0 and >5.0. The printed categories leave a gap between 1.0 and 2.0, which is noted.
  - Total eggs and meat: <9, 9-11, 11.5-14.0, 14.5-18.0 and >18.
  - All are recorded correctly.

## Case-count re-derivation
Every one of the 8 exposure sets sums to 212, which equals the paper's total colon cases. Every reference category is 1.00. The age-adjusted rows have no CIs, which is correct because none are printed. Every multivariate point lies inside its CI.

## Models and flags
- m1 (age-adjusted): Mantel-Haenszel in five-year age categories (p. 40). Correct.
- m2 matches Table 3 fn a: age, total energy, height, parity, total vitamin E, vitamin E × age, and vitamin A supplements.
  - adj_energy = yes (covariate) is correct.
  - All other flags = no is correct. Smoking, BMI, alcohol and the other variables were only "considered" and are not in the final list.
- P for trend: the paper prints "χ² (P)". The database records the P value, e.g. 0.62 and 0.69 for main dish, which is correct.
- Source pages are correct: p. 43 for the meat rows, p. 44 for total eggs and meat.

## studies.csv
Confirmed:
- Population: n = 35,216 at baseline (abstract: 35,215, as noted); 212 colon cases; 78 rectal cancers not counted as cases.
- Diet: 127-item FFQ; diet_repeated = no.
- Outcome via the SEER registry; age range 55-69.

## Missing estimates
None. Every meat-type line in Table 3 is coded. Liver is mentioned only in the Discussion as showing "no relationship", with no numbers.

## Rows confirmed correct
All 74 numerically. Every field is confirmed except the main-model flag on 013-016 / 021-024 and the exposure value on 065-074 (see above).

## Internal inconsistencies in the paper
- The abstract gives n = 35,215; the Methods give 35,216.
- The Discussion says risk was suggested to rise "to a lesser extent" with main dish. The multivariate RRs are 1.14, 0.87 and 1.21 (P trend 0.69), with no gradient.
- White-meat category labels skip from <1.0 to 2.0, as printed.

## Verdict
Fit for use. All numbers are correct. Fix the duplicate poultry main sets before running any negative-control pools.