# K10 control – fraser1999 (Fraser, Am J Clin Nutr 1999;70(suppl):532S-8S, Adventist Health Study 1 – review/summary)

Sources: historisk/fraser1999.txt and a pdftotext extraction of fraser1999.pdf (native text layer). Pages come from the running heads: Table 7 is on 536S, Table 8 on 537S, and the legume text on 535S.
Rows checked: 8.

## Discrepancies
| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| fraser1999_001 | exposure / exposure_includes_processed | total_meat / unclear | the paper defines nonvegetarians as eating "meat, fish, or poultry" ≥1 time/wk in total, against vegetarians eating none (Table 1–2 fns, p.533S). So the exposure includes fish, as the exposure_text already states. Add a note that it is flesh food incl. fish. | minor |
| fraser1999_002–007 | is_main_model | yes | These are subset analyses (restricted to white meat <1/wk, or red meat <1/wk) in a secondary review paper whose covariates are not stated. The primary paper singh1998 codes the same subsets (Table 4) as `no`, and its whole-cohort Table 3 as main. For consistency they should be `no`. Otherwise a "main-model" filter picks up two conflicting AHS-1 versions. | major (flag) |
| fraser1999_005 | row existence | reference 1.0 "Never", added | Table 8 prints no "Never" row for white meat. The structural reference is reasonable (the paper says RR is relative to never consumers only in Singh) but it is not printed. The note already says so. | minor |
| studies.csv fraser1999 | cases_colon | 107 | Table 7 "Incidence n" is 107 for the vegetarian vs nonvegetarian comparison only. Table 8 subsets have 112 and 82 cases, and Singh 1998 reports 157 colon cases for the full cohort. The note explains this, but cases_colon=107 will fail check 5 for the Table 8 sets. | minor |

No numeric discrepancies. Every number matches the paper: 1.88 (1.24, 2.87), P 0.0032; Table 8 1.37 (0.85–2.20), 1.86 (1.15–3.02), 1.50 (0.94–2.41), 3.00 (1.45–6.20), P trend 0.01 (fn 2); legume-stratified 2.68 (1.24, 5.78).

## Cross-study point: Fraser 1999 Table 8 vs Singh & Fraser 1998 Table 4 (confirmed discrepancy)
Both papers describe the same analysis: AHS-1, the same subsets, the same case numbers (112 and 82), and the same 136,864 / 100,649 person-years in Fraser fn 1. The numbers differ:

| subset / category | Fraser 1999 Table 8 | Singh 1998 Table 4 |
|---|---|---|
| red meat >0–<1/wk (white <1/wk) | 1.37 (0.85–2.20) | 1.40 (0.87–2.25) |
| red meat ≥1/wk | 1.86 (1.15–3.02), P trend 0.01 (Wald) | 1.90 (1.16–3.11), P trend 0.02 |
| white meat >0–<1/wk (red <1/wk) | 1.50 (0.94–2.41) | 1.55 (0.97–2.50) |
| white meat ≥1/wk | 3.00 (1.45–6.20), P trend 0.01 | 3.29 (1.60–6.75), P trend 0.006 |

I verified the Singh values against Understudier WCRF/singh1998.txt, Table 4 (p.767). Singh's text also gives a "further categorization" with red >0–<1/wk at 1.38 (0.86–2.20), close to Fraser's 1.37 (0.85–2.20). This suggests Fraser reports an earlier or alternative model run, but no covariate list is printed. Likewise, Fraser's legume-stratified red-meat RR 2.68 (1.24–5.78) matches neither Singh Fig 1 (nonvegetarian at legume <1/wk: 2.54, 1.20–5.37) nor Singh Fig 2 (2.28, 1.28–4.05).

**Conclusion:** Singh 1998 is the primary, peer-reviewed source with a stated model. Fraser 1999 should be treated as a secondary report of the same cohort. It must never be pooled with Singh (same cohort_id `ahs`, which is in place). Singh should be preferred for any AHS-1 row. The database notes already record the discrepancy correctly in rows 002–008.

## Missing estimates
None. Beef and IHD/bladder results are not colorectal.

## Confirmed
Numeric values in all 8 rows; source pages; adjustment m1 (Table 7 fn: age and sex). m2 is correctly all "unclear", since the paper only says "On multivariate analysis (15)".

## Verdict
Numerically correct, but it is a secondary source. Downgrade rows 002–007 to non-main and use singh1998 for AHS-1.