# K10 control – goldbohm1994 (Goldbohm et al., Cancer Res 1994;54:718-23, NLCS)

Source: 23_Goldbohm1994.txt (OCR of the Maastricht repository PDF). Tables 5 and 6 are legible apart from isolated characters. No PDF image checked.
Rows checked: 57.

## Discrepancies
| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| goldbohm1994_001–015, _040 | exposure | total_meat (includes_processed = no) | "Fresh meat (including chicken)": all meat except processed meat, plus chicken (Table 5 fn c; Methods p.719). This is not all meat, because processed meat is excluded. The vocabulary has no exact code. Note "fresh meat incl. poultry, excl. processed" explicitly, and exclude these rows from any total_meat pooling that assumes processed is included. | major (exposure definition) |
| goldbohm1994_011–015, 024–027, 036–039 | n_cases | empty | not printed for both sexes, but they are the men + women sums: fresh 44/41/56/40/34; processed 23/74/59/59; fish 70/53/33/59 (each sums to 215) | minor (could be filled, derived) |
| studies.csv goldbohm1994 | red_meat_definition | "all meat except processed, plus chicken" | correct, but there is no red-meat aggregate at all; the closest true red-meat variables are beef/pork/minced in Table 6 | note |

No numeric discrepancies.

## Rows resting on OCR correction
| estimate_id | correction | supported by prose? |
|---|---|---|
| _021 | women, processed 0–10 g: "5.22" → 1.22 | **No prose**. The geometric mean of the CI (0.66–2.26) is 1.221, which supports it strongly. UNSUPPORTED BY PROSE, near-certain. |
| _030 | men, fish 10–20 g: cases "It" → 11 | **No prose**. The sum check (34+28+11+32 = 105 men) proves it. Accept. |
| _034 | women, fish 10–20 g: "114" → 1.14 | **No prose**. CI geometric mean is sqrt(0.66×1.97) = 1.14. UNSUPPORTED BY PROSE, near-certain. |
| _043 | minced meat: mean/SD "6 18 {7" → 18/17 | **No prose**. The "6" is the model letter "b" (pork prints as "h"). The sum of fresh-meat type means, 25+38+18+2+14+3 = 100, matches the total fresh meat mean of 99. Accept (only mean/SD in notes). |
| _051 | cooked liver mean "i" → 1 | The sum of processed means, 4+1+3+1+5 = 14, matches the total of 14. Accept. |
| _047 | 1.17 (1.03–1.33) | **Yes**: abstract and Results p.721. |
| _027 | 1.72 (1.03–2.87), P 0.02 | P supported by text ("Only processed meat showed a significant (P = 0.02)"). |
| _019/_023 | P trend 0.06 / 0.10 | **Yes**: text p.720. |

## Missing estimates
None. Table 6 has 13 rows, all present. The frequency RRs in the text (0.65, 0.56, 0.78, 0.81) are present with point estimates only, as printed. Tables 3–4 (fats, meat fat, meat protein) are nutrients, so they are out of scope. The models with fat from meat added give no numbers.

## Confirmed
All RR/CI/P values and case counts in rows 001–057. The case sums are men 105 and women 110 for every Table 5 exposure, matching the text. Medians, models m1–m6, pages (Table 5 p.721, Table 6 p.722, frequency text p.720) and studies.csv are confirmed: 120,852 aged 55–69; subcohort of 1525 + 1598 = 3123 analysed; 3.3-year follow-up with the first year excluded; 215 colon cases; Spearman r for fresh meat 0.46.

## Internal consistency
Consistent. Text and Table 5/6 agree (1.17 per 15 g; P 0.06/0.10/0.02).

## Verdict
Fit for use. The one substantive point is that "fresh meat" includes chicken and excludes processed meat. It must not be used as red meat, nor as total meat including processed. Two point estimates (_021, _034) rest on a decimal/digit restoration that the CI confirms but the prose does not.