# K10 control – willett1990 (Willett et al., NEJM 1990;323:1664-72)

Source used: rendered PDF pages (pdftoppm -r 200) of 21_Willett1990.pdf (Drive [Drive-id]), pp. 1664-1669. Every row checked against the page image.
Rows checked: 42.

## Discrepancies

Table 3 (p. 1668) prints bracket lines spanning merged frequency columns. With the brackets visible the column assignment is unambiguous:
- Beef/pork/lamb main dish: `<1/mo` | `1-3/mo + 1/wk` merged (42 cases) | `2-4/wk` | `5-6/wk` | `>=1/day`
- Processed meats: `<1/mo` | `1-3/mo` | `1/wk` | `2-4/wk` | `5-6/wk + >=1/day` merged
- Chicken without skin: `<1/mo` | `1-3/mo` | `1/wk` | `2-4/wk through >=1/day` merged
- Fish: `<1/mo` | `1-3/mo` | `1/wk` | `2-4/wk` | `5-6/wk + >=1/day` merged

| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| willett1990_002 | category_label | 1-3/mo (column inferred) | 1-3/mo and 1/wk merged (bracket), Table 3 p.1668 | major (dose) |
| willett1990_003 | category_label | 1/wk (column inferred) | 2-4/wk | major (dose) |
| willett1990_004 | category_label | 2-4/wk or wider (alignment unclear) | 5-6/wk | major (dose) |
| willett1990_005 | category_label | >=1/day | >=1/day – correct; note-text on alignment can be removed | – |
| willett1990_006 | category_label | lowest frequency (<1/mo assumed) | <1/mo (confirmed) | minor |
| willett1990_007 | category_label | category 2 of 5 | 1-3/mo | minor (dose recoverable) |
| willett1990_008 | category_label | category 3 of 5 | 1/wk | minor |
| willett1990_009 | category_label | category 4 of 5 | 2-4/wk | minor |
| willett1990_010 | category_label | category 5 of 5 (highest) | >=5/wk (5-6/wk + >=1/day merged) | minor |
| willett1990_012/013 | category_label | "column inferred" | 1-3/mo and 1/wk confirmed | minor (remove "inferred") |
| willett1990_014 | category_label | >=2/wk (text) | 2-4/wk to >=1/day merged = >=2/wk, confirmed | – |
| willett1990_015–019 | category_label | lowest / category k of 5 | <1/mo, 1-3/mo, 1/wk, 2-4/wk, >=5/wk | minor |
| willett1990_024–028 | n_cases | empty | 25, 27, 28, 26, 44 (sum 150) Table 4 p.1668 | major (missing counts; GL needs them) |
| willett1990_029–033 | n_cases | empty | 38, 29, 39, 19, 25 (sum 150) | major |
| willett1990_034–038 | n_cases | empty | 22, 26, 27, 29, 46 (sum 150) | major |
| dose_conversions willett1990_c1–c4 | notes | "column alignment partly lost" | alignment recoverable from PDF; frequency bounds above can be filled | minor |

The OCR lost the Q3 value in each Table 4 cases row (28, 39, 27). All three rows now sum to 150 = total colon cases, which confirms the reading.

All numbers (cases, person-years, RR, CI, chi/P for trend) in rows 001–038 match the page image exactly. Text estimates 039 (2.44, 1.44–4.14, p.1668), 040 (1.61, 1.03–2.53, p.1669), 041 (0.63, 0.39–1.00, p.1669) and 042 (colon+rectum ratio 2.30, 1.48–3.56, P trend <0.0001, p.1669) match.

## Missing estimates
None. Rectal cancer (39 cases) is reported only as "P for trend = 0.08" for the ratio, with no RR. Chicken with skin has P trend 0.65 only, with no RR. Neither can be extracted.

## Confirmed
All numeric fields in rows 001–042. Model/adjustment assignment (m1 age-adjusted Table 3 with food frequencies not energy-adjusted; m2 Table 4 age-adjusted with energy-adjusted intakes; m3 mutual model in text) is consistent with the titles and footnotes. Exposure coding is correct: red_total includes hot dogs, preserved meats and bacon (Table 4 fn †), and red_main_dish is "partial" (fn: steak, roast, or ham). studies.csv fields are correct: 88,751 women aged 34–59, 1980 FFQ with 61 foods, 512,488 person-years, 6 years of follow-up, 150 colon cases (90% pathology-confirmed), 39 rectal cases.

## Internal consistency
Text and tables agree: daily main-dish RR 2.49 (1.24–5.03) and ratio Q5 2.49 (1.50–4.13) are each repeated in the abstract. Table 4 has 292 women excluded (fn), so its person-years differ slightly from Table 3, which is expected.

## Verdict
Fit for use once the corrections are made. The numbers are all right. The main-dish dose labels for ranks 2–4 are shifted and must be corrected before any dose-response use, and the Table 4 case counts should be filled in (25/27/28/26/44 etc.).