# K10 control – chen1998 (Chen et al., Cancer Res 1998;58:3307-11, Physicians' Health Study nested case-control)

Sources: historisk/chen1998.txt; Chen1998.pdf from Drive, with PDF pp.4–5 (journal pp.3309–3310, Tables 2–3) rendered at 200 dpi and read cell by cell.
Rows checked: 42.

## Discrepancies
None. Every case count, control count, RR/OR, CI and P trend in Table 2 (rows 001–021) and Table 3 (rows 022–042) matches the rendered page images. This includes the cells cut off in the txt: _012, _015, _018, _021, _033, _036, _039, _040, _042, and _029's upper CI of 4.47.

| item | note | severity |
|---|---|---|
| effect_measure | Table 2 prints "RR" and Table 3 prints "OR" (fn b). Both come from the same conditional logistic regression, so they are odds ratios. The database follows the printed labels, which is acceptable under "as published". | none/minor |
| _022–_042 | the text says 107 cases / 116 controls; the Table 3 controls sum to 44+53+18 = 115. This is the paper's own inconsistency, already noted in the database. | – |

## Re-derivations
- Table 2, all men: cases 62+103+43 = 208 of 212 (4 missing red-meat data), controls 66+112+39 = 217 of 221.
- Each genotype stratum sums to ≤ 208/217.
- Medians 0.27/0.77/1.46 servings/day (Methods p.3308) are correctly used as dose_point with paper_median.
- The model (matched on age and smoking; plus BMI, physical activity and alcohol) agrees with the Methods.

## Missing estimates
None. The colon-only RRs in the text (0.92 and 0.77) are for genotype, not meat. For chicken and fish the text says only "not related", with no numbers.

## Confirmed
All 42 rows, adjustment set m1, and studies.csv (nested_cc; 212 cases / 221 controls; 169 colon; 13 years of follow-up; red meat includes hot dogs, so includes_processed = yes).

## Verdict
Fit for use. The only main-model rows are 001–003 (three categories, servings/day). The genotype and age strata are correctly marked non-main.