# K9b – islam2019 (opus-S2 rows, 788) vs Supplementary Tables S1–S4

Source: /home/claude/drive/supplements/islam2019_S1-S4.txt. The txt was cross-checked against the source docx: all 1,251 HR and case/person-year tokens identical.
Method: the brief asked for a systematic 25% sample plus all continuous rows. **All 788 rows were compared by script** (point, CI, n_cases, person_years, n_in_category, P-trend on the top category, sex by table, model by label, completeness). The 100 continuous rows are included. Labels, dose fields and flags were checked by tabulating them across all rows; notes were read for every flagged row and a sample of rows.

## Discrepancies
| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| adjustments islam2019_m5 | notes | "Only one estimate from this model is printed (beef, distal colon, women, text)" | 344 S2 rows now use m5 (age/area-adjusted rows in S1–S4) | minor (stale) |
| none numeric | – | – | – | – |

## Checked and confirmed
- S1/S2 (frequency; JPHC I/II, JACC, MIYAGI, OHSAKI): beef, pork, processed red meat, chicken × CRC, colon, rectum (4 categories) and proximal/distal (3 categories; "3-4 times/week or more" = combined per footnote 1). Age/area model (m5) and excluding-<3-y model (m1/m2) all correct, including N, cases, PY and P-trend.
- S3/S4 (quantity; JPHC-5y, TAKAYAMA): unprocessed red meat, beef, pork, processed red meat, chicken × 5 outcomes × quartiles + continuous (per 100 g; processed per 50 g, footnote 2). All 100 continuous rows correct. Quartile cut-points are not printed, so empty dose fields are correct.
- Excluding-<3-y rows use the sex- and pool-specific multivariable sets m1–m4 (same footnote 3). OK.
- is_main_model=no for all S2 rows: correct; the multivariable rows are the opus-E main-table rows.
- Print errors recorded as printed and flagged: 455 "0.91 (0.81-0.02)", 1051 "1.07 (0.82-0.40)", 502, 1072, 1161 (asymmetric CIs). Agree with all flags.
- Extra check: the 394 opus-E main-table rows (Tables 2–5) equal the supplement's multivariable rows exactly, including typo row 261 "0.58 (0.70-1.04)". Only P-trend formatting differs (".99" vs "0.99").

All 788 rows confirmed.

## Missing estimates
None. Every age/area-adjusted and excluding-<3-y cell in S1–S4 has a row.
Enhancement (not an error): the opus-E main-table rows for the quantity pool (e.g. 145–149) have empty n_cases/person_years. S3/S4 print them (identical populations), so they can be filled.

## Paper-internal inconsistencies
- Print errors listed above (point outside CI in S1 line 49, S4 line 513; S3 line 423 multivariable; missing parenthesis S4 line 593).