# K1 control – ollberding2012 (MEC, Int J Cancer 2012, Early View)

Controller: K1, 2026-10-30. Sources read: `ollberding2012.txt` (pp. 1–9) and `ollberding2012_supplement.txt` (Suppl Table 1). Database files only; no study card or work files were read.

## Method
- A script parsed Table 2 (p. 5) and Suppl Table 1. It compared every numeric field on all 120 rows: quintile median, cases, n at risk, point, CI and p_trend for Table 2; label/cut-points, cases, n at risk, point, CI and p_trend for the supplement. That is 1,120 field comparisons, with **0 mismatches**.
- Definitions, models, flags and adjustment sets were checked by reading the methods (pp. 2–3) and the Table 2 and Suppl Table 1 footnotes.
- Case sums were re-derived independently. Every Table 2 set sums to 3,404 cases / 165,717 at risk, which matches the abstract and p. 2. Each ethnic subgroup set sums identically across the four exposures (AA 648, JA 1,240, Latino 613, White 681). The 222 missing cases are Native Hawaiians, whom the supplement excludes.
- Every Table 2 median lies inside the Suppl cut-point interval used for `dose_low`/`dose_high` (all 20 checked).

## Discrepancies

| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| (studies.csv) ollberding2012 | cases_colorectal | empty | 3,404 invasive CRC (abstract; p. 2) | major (check 5 has no total to compare against) |
| (studies.csv) ollberding2012 | cases_colon / cases_rectum | empty | not reported (subsite results "data not shown", p. 6) | none – empty is correct |

No row-level discrepancies.

## Key points verified
- **exposure_includes_processed:**
  - "Total red meat" → `red_total`, `yes`. This is correct by inference: the paper reports a separate "Red meat excluding processed red meat". No food list is given.
  - "Red meat excluding processed red meat" → `red_unprocessed`, `no`. Correct.
  - Total meat `unclear` and Processed `yes` are correct.
- **is_main_model:** `yes` only on the Table 2 "Multivariable model" sets (rows 006–010, 036–040, 066–070, 096–100). This is correct: the abstract reports this model. The basic model and the ethnic strata are `no`.
- **CUP-relevant rows.** CUP's contrast "47.99 vs 7.41 g/1000 kcal/day" is the Total red meat quintile medians. CUP therefore used rows **036–040** (MV, red_total). The rule-conform (unprocessed) set is **066–070**: Q5 0.98 (0.87–1.10), p trend 0.584. Both sets are correctly coded.
- **Doses:** in g/1,000 kcal/day (energy density, p. 3). The medians are `paper_median`. No g/day values exist in the paper, so `g_day_assigned` is correctly left empty. Any per-100 g conversion needs an external energy assumption.
- **Denominator:** "Cases/no. at risk" is coded `n_participants`. Correct; person-years are not printed.
- **adjustments m2:** the flags match the Table 2 footnote 4 covariates. `adj_energy = yes` (log kcal in the model) with `energy_method = density` is right. m1 and m3 also match.
- **source_page:** Table 2 = p. 5 (Early View pagination); supplement = suppl p. 1. Correct for the file at hand.

## Missing estimates
- None that are required. Colon/rectum, continuous, lag and adenocarcinoma-only analyses are all "data not shown" (p. 6), and no sex-specific estimates are printed.
- **Optional, a scope decision:** "Total meat cooked dark brown" (Table 2: 5 quintiles × 2 models, in the HCA subcohort, 1,757 cases) plus Suppl Table 1 (4 ethnic groups × 5) give 30 rows. It is a meat exposure (a doneness subset of total meat) and arguably within R1. The extractor excluded it deliberately and documented this.
- Out of scope and correctly excluded: the meat-and-fat dietary pattern, HCAs, pack-years, and Table 3.

## Rows confirmed
All 120 rows (ollberding2012_001–120) are confirmed correct.

## Internal inconsistencies in the paper
- The methods (p. 3) list multivitamin use but not HRT. The Table 2 footnote lists HRT ("females only") but not multivitamin. The extractor noted this; confirmed.
- Suppl Table 1 typos are outside the coded scope: PhIP AA Q4 "1.20 (0.85-1.07)"; PhIP Latino Q5 "10/9470"; MeIQx cut-point gap (33.9 → 40.0).
- Suppl AA processed Q2 is printed "1.1 (0.85-1.44)" with one decimal. Row 102 records it as printed.
- Table 1 Q1/Q3/Q5 cases and at-risk counts match Table 2 exactly.

## Verdict
Fit for use. The numbers are exact. The only gap is the empty `cases_colorectal` (3,404) in studies.csv.