# K5 control – papier2025 (Million Women Study diet-wide analysis, Nat Commun 2025;16:375)

Controller: K5 · 2026-10-30 · Full text: `post2015/papier2025.txt` (PDF text; Supplementary Data 1–6 and Supplementary Methods not available)
Method: I read Table 3 (p. 7), Results (p. 2), Discussion (p. 8), Methods (p. 9) and the Fig 1–3 legends. I compared all 5 rows by hand and grepped the whole text for every meat, poultry and fish mention.

Rows checked: 5.

## Discrepancies

| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| papier2025_001 | is_main_model (with is_calibrated = yes) | yes | KODEBOK 4.2(b) excludes calibrated models from `is_main_model`. Elsewhere in the database, re-measured-mean trend rows are `is_calibrated = yes, is_main_model = no` (knuppel2020 105 rows, bradbury2020 30, norat2005 41). Papier is the only study with a calibrated row set to main. | major (rule conflict). Defensible, because the paper prints no uncorrected estimate, but it must be a documented exception, or pool.py's main-model selection will treat Papier differently from Knuppel/Bradbury. |
| studies.csv papier2025 | baseline_start / baseline_end | 1996 / 2001 | These are recruitment years. The dietary baseline is the first resurvey, "median year 2001, IQR 3" (Methods). | minor (the notes explain this) |

## Specific checks requested
- **Unit:** "Red/processed meat | 30 g/day" (Table 3 Trend increment). `dose_unit = per_30g` is correct. The increment is on the WebQ re-measured scale (Methods); the notes state this.
- **Values, Table 3, red/processed meat:**

  | Model | Paper | Database row |
  |---|---|---|
  | Main | 1.08 (1.03,1.12) | _001 |
  | + calcium | 1.06 (1.01,1.11) | _002 |
  | + dairy milk | 1.07 (1.03,1.12) | _003 |
  | + fruit | 1.06 (1.02,1.11) | _004 |
  | + wholegrains | 1.06 (1.02,1.11) | _005 |

  All match.
- **Calibration:** the trend is computed from "re-measured mean intakes for each baseline category" (Table 3 fn 1; Methods p. 9), to "reduce the impact of regression dilution bias". `is_calibrated = yes` is correct and follows the convention used for knuppel2020 and bradbury2020. It is not EPIC-type regression calibration, and the notes say so.
- **p_value "<0.01"** comes from the Results p. 2 text. Correct.
- **Exposure:**
  - `red_total`, "Red and processed meat", `includes_processed = yes`. Correct.
  - Components are defined only in the Supplementary Methods, so `red_meat_definition` correctly says so.
- **Cases and n:** 12,251 cases and 542,778 women. Correct (title, abstract and Table 1). Mean follow-up 16.6 y. Correct (Table 1, abstract).
- **Adjustment sets m1–m5:**
  - Stratification on year of birth, date of survey and region gives `adj_stratified_center = yes`. Correct here, because this is true stratification.
  - Covariates: deprivation, education, BMI, height, strenuous exercise, energy fifths, alcohol, smoking, HTM, family history. Flags correct.
  - m2 adds calcium. m3 adds dairy milk (no calcium flag). m4 adds fruit (`fruit_veg = yes`, noted). m5 adds wholegrains (fiber = no). All correct.
- **Source pages:** Table 3 lies between the p. 6 and p. 7 page footers, so p. 7. Correct.
- **CI symmetry:** _001 "1.08 (1.03,1.12)" is slightly asymmetric on the log scale (0.23). That is consistent with two-decimal rounding of a narrow CI. Not an error.

## Missing estimates
- **None in the available text.** The paper says separate red meat and processed meat results exist ("larger association for processed meat than for red meat") but gives no numbers. These, plus category-level RRs (Fig 3 is graph only) and subsite results, are in Supplementary Data 1/4, which is not available.
- **"Equivalent to a 29% higher risk per 100 g/day"** (Discussion p. 8) has no CI and is a rescaling of _001. It is correctly not coded as a separate row.
- **Gap for use:** the only row is red+processed combined. Without Supplementary Data 1 this study gives **no unprocessed red meat estimate**.

## Confirmed rows
papier2025_001–005: all numbers confirmed. Only the is_main_model rule issue on _001 applies.

## Internal inconsistencies in the paper
- None affecting meat.
- Table 3 has printing errors in other rows: magnesium + calcium "0.83 (0.83,1.01)" and magnesium + wholegrains "0.80 (0.99,4.69)". These are not meat rows, but they show Table 3 was not proofread cleanly.
- The p < 0.01 for red/processed meat is not stated as raw or FDR-corrected. The notes say this.