# K5 control – mehta2020 (Sister Study, Cancer Epidemiol Biomarkers Prev 2020;29:141–150; author manuscript)

Controller: K5 · 2026-10-30 · Full text: `post2015/mehta2020.txt` (NIH author manuscript; Supplemental Tables 1–5 not available)
Method: Table 2 (130 rows) and Table 5 (100 rows) parsed by script. Every case count, median, HR, CI and p-trend was compared with estimates.csv. The 4 text rows were read by hand. dose_conversions was compared with the category medians.

Rows checked: 234.

## Discrepancies

| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| mehta2020_131–230 (Table 5 category rows) | dose_point_source | paper_median | Labels read "Tertile 2 (11.20 g/day)". Footnote b says trend uses tertile medians, but the label value is not called a median. | minor (assumption is noted) |

No critical or major discrepancies.

## Specific checks requested: "N (N adjusted)"
- Table 2 prints cases as "216 (210)" and per quartile "51 (51) | 53 (52) …".
- In every one of the 13 exposures, the database puts the first number on Crude rows and the parenthesised number on Adjusted rows, both on category rows and continuous rows (216 vs 210). Verified cell by cell by script: 0 mismatches.
- Sums: every Crude set sums to 216 and every Adjusted set to 210.
- Table 5 "N" = cases in each stratum. Every stratum sums to the continuous-row N, except the paper's own inconsistency below.

## Other checks
- **Point, CI and p-trend:** 0 mismatches in 230 table rows.
- **Text rows verified (txt l.213):**
  - 231: organ-meat-adjusted Q4, 1.02 (0.66‐1.58).
  - 232: shellfish, 0.87 (0.60‐1.29).
  - 233: nutrient density, red meat, 1.17 (0.78‐1.77).
  - 234: nutrient density, bacon, 1.22 (0.81‐1.82), p-trend 0.32.
- **Units:** continuous increments match the Table 2 symbols (+ 50 g, # 25 g, ^ 10 g). Unprocessed red meat and total/white meat are per_50g; processed and the subtypes per_25g; bacon, hot dogs and breakfast sausage per_10g. Ham/lunch meat and hamburgers are per_25g. All correct.
- **Exposure definitions:**
  - Unprocessed red meat = FPED beef, veal, pork, lamb and game, plus organ meat. Coded `red_unprocessed`, `no`. Correct.
  - Beef/pork `includes_processed = unclear` is defensible (FFQ items, not defined).
  - Hamburgers → `red_unprocessed`. Other products → `processed`. Correct.
- **Models:**
  - m2 matches Table 2 fn a (kcal, BMI, education, MET-h, race, family history).
  - m3 (BMI dropped) is used for the BMI-stratified columns, per Table 5 fn a. Correct.
  - Age is the time scale, so `adj_age = yes`.
  - Sensitivity covariates are not flagged, because no estimates were printed for them. Correct.
- **is_main_model** = yes only on Adjusted rows of the 8 aggregate/subtype exposures. Consistent with KODEBOK 4.2.
- **Reference rows:** 1.0 with empty CI.
- **dose_conversions:** all 76 mehta/gilsing rows were checked against the category medians. 0 mismatches for mehta.
- **studies.csv:**
  - n 48,704; 216 CRC; rectum 30. Correct.
  - Median follow-up 8.7 y. Correct; the 4.8 y in Results is the cases' median time to diagnosis.
  - 420,574.35 person-years in notes. Correct.
  - Colon total not printed, so left empty. Correct.

## Missing estimates (scope-dependent)
The extractor left these out, citing a brief that limits cooking/doneness to "same table". KODEBOK R1 ("alle estimater for rødt kjøtt (alle definisjoner)") arguably covers Table 3:
- **Table 3:** red meat intake by cooking practice, quartiles + per 50 g. About 22 rows:
  - grilled red meat
  - grilled red meat excl. pork
  - pan-fried (Q1–3 combined)
  - medium-well to charred
  - grilled and medium-well to charred (Q1–2 combined)

  Examples: grilled red meat per 50 g is 1.20 (0.83‐1.73); grilled excl. pork per 50 g is 1.35 (0.90‐2.01).
- **Table 4:** preferred cooking method and doneness per product (steak, hamburger, pork chops, bacon/sausage) vs "don't eat". About 36 rows. These are preference categories, not intake doses, and are less clearly in scope. Example: grilled/barbequed steak 2.23 (1.20‐4.14).
- **Supplemental Tables 1–5** (colon/rectum separately, confirmed cases only, mutual adjustment, exclusions) are not available.

Recommendation: decide the scope for Table 3 explicitly. If in scope, add it as `red_unprocessed` with an exposure_text naming the cooking method, `is_main_model = no`.

## Confirmed rows
All 234 rows confirmed (mehta2020_001–234).

## Internal inconsistencies in the paper (correctly recorded as printed)
1. **White meat, adjusted Q2:** "1.12 (0.77-1.79)". The CI is identical to the adjusted Q4 CI and asymmetric around 1.12. Probably a copy error. Flagged in the database.
2. **Table 5, hot dogs/dinner sausages, age 55+:** tertile cases 62+35+50 = 147, but the continuous N is 148.
3. **Table 5 age-group Ns** sum to 48,706, not 48,704.
4. **Table 3:** broken parentheses in CIs, e.g. "(0.91-1.88" and "0.87-1.92)" (not coded).
5. **Follow-up:** Results say "median 4.8 years of follow-up", abstract and Discussion say 8.7 y. The Discussion resolves this: 4.8 y is time to diagnosis for cases.