# K10 control – sellers1998 (Sellers et al., Cancer Causes Control 1998;9:357-67, Iowa Women's Health Study; Drive file bazyk1998.pdf)

Sources: historisk/sellers1998.txt and a pdftotext extraction of bazyk1998.pdf (native text layer). Table 3 is on journal p.362, between the 361 and 362 footers.
Rows checked: 48.

## Discrepancies
| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| sellers1998_013–024 | exposure_includes_processed | no | red meat = "liver, hamburger, beef stew, beef, and venison" (Table 3 fn g). Under codebook R6 (the hamburger rule), "unprocessed" that includes hamburger must be coded `partial`. It also contains liver (organ). | major (flag) |
| studies.csv sellers1998 | includes_processed | no | should follow R6: partial | major (flag) |
| _001–_012 | exposure_includes_processed | unclear | Total meat is not defined in the paper. The baseline means (13.5 servings/wk total vs 5.9 red + 3.5 white + 1.9 nitrate = 11.3) show it is not the simple sum of the three groups. "unclear" is correct. | – |
| cohorts.csv iwhs | notes | "Bostick 1994; Lee 2004 (HvL)" | sellers1998 is also in iwhs but is not listed in the cohort note. The cohort_id is correct in both studies and estimates, so pooling exclusion works. | minor |

No numeric discrepancies. All 48 rows (cases, age-adjusted RR, multivariate RR, CI, trend P, by family-history stratum) match Table 3 exactly.

## Re-derivations
- Case sums: every exposure gives 180 in the FH-negative stratum and 61 in the FH-positive stratum, total 241 = the paper's total colon cases.
- The trend P is placed on the top-category multivariate row only. This is correct.
- The age-adjusted rows have no CI printed. This is correct.
- The model fn b reads "age at baseline, total energy intake, and history of rectal colon polyps", matching m2.

## Missing estimates
None. The paper gives no estimate across both strata combined; meat results are stratified by family history only (text p.360: no clear patterns for meat). Rectal cancers (78) are not analysed.

## Cross-study point: Sellers 1998 vs Bostick 1994 (same cohort)
Both studies have `cohort_id = iwhs`, and both studies.csv notes say they must never be pooled together. Bostick 1994 is follow-up to 1990 with 212 colon cases and the "beef, pork, or lamb as main dish" exposure. Sellers 1998 is follow-up to 1995 with 241 colon cases, an aggregate red meat including liver and hamburger, and estimates only within family-history strata. The exposure definitions are not comparable, and Sellers has no whole-cohort estimate. **Sellers should not replace Bostick as the IWHS row in pooling.** It could only enter through a fixed-effect combination of the two strata, which would be a derived estimate. Confirmed that no row in Sellers is is_main_model across both strata (there are none).

## Confirmed
All numeric fields in rows 001–048; source page 362; adjustment sets m1/m2; studies.csv fields: 35,216 women aged 55–69, follow-up to 31 Dec 1995, 241 colon cases, 127-item FFQ, registry ascertainment.

## Verdict
Fit for use after the R6 flag change (red meat → partial). It has limited value for pooling (stratified only, servings/week, same cohort as Bostick 1994).