What the analysis says
The cluster asks whether the egg-CVD association is genuinely modified by region/population or the apparent heterogeneity (above all the protective Asian signal) is an artefact. Members: H-7 - Inverse egg-CVD association in China Kadoorie reflects healthy-user confounding not causation confounding, H-22 - Region-population modifies the egg-CVD association (inverse in Asia, null in West) genuine modification, H-50 - Some other origin of regional heterogeneity residual (exposure-range/measurement/chance). Prior [0.467, 0.105, 0.428] → posterior [0.370, 0.044, 0.586]: genuine biological modification nearly ruled out, the residual now the plurality. The exposure-range reframe did most of the work — E-128 - O-8 × HC-2 — Asian inverse measured only below 1 egg per day (Kadoorie spans only 0.29-0.76 egg/day), the within-Asia U-shape, and the West’s sparse high-intake data all anchor H-50=1, recasting “inverse in Asia” as “inverse at low intake on one universal dose-response.” PURE’s cross-region consistency (CG-7 - HC-2 joint over O-86+O-87+O-88+O-84) and KoGES’s non-replication of the inverse (E-126 - O-72 × HC-2 — second Asian cohort shows no egg-CVD inverse) scored H-22 at 0.45-0.5, sinking it.
What the model may not capture
H-7 and H-50 are not cleanly exclusive: both are “not genuine modification,” and the model itself notes (E-122 - O-48 × HC-2 — region interaction non-significant, heterogeneity persists within regions) that O-48 “barely separates H-7 from H-50.” The robust conclusion is the pooled 0.956 on “not genuine biological modification”; the 0.370/0.586 split between confounding and exposure-range is softer than it looks and partly a labelling choice. More important: is the answer on the list? The members carve “region as biological effect-modifier” against “artefact,” but a fourth reading — the gradient is real and systematic yet driven by background-diet contingency (what eggs displace/accompany, HC-10) rather than population biology — is not squarely any member. E-131 - O-17 × HC-2 — same region gradient significant for T2D is the load-bearing tension: the same US-harm/Asia-protective gradient reproduces on an independent outcome (T2D, P-interaction 0.01), which pure chance cannot produce and which genuine systematic modification predicts — yet A-5 reads it as shared confounding-by-Western-pattern, and it lands as 0.85 for H-7. That a reproducible cross-outcome gradient still leaves H-22 at 0.044 is the point most worth doubting: it may be under-weighting a real, non-biological but non-artefactual regional structure.
What would help
- Egg-CVD estimates within Asian and Western cohorts over a common intake range, to tell dose-position from region — does not exist (Western high-intake data too sparse, per O-42).
- An RCT or Mendelian-randomisation contrast across populations — does not exist.
- Whether the T2D and CVD region gradients survive adjustment for the same background-diet confounders — exists, unread.
Confusions and contradictions
Two evidence streams point opposite ways and the analysis resolves rather than reconciles them: the reproducible cross-outcome region gradient (E-131) reads as real modification, while PURE’s cross-region consistency and the exposure-range facts read as one universal curve sampled at different doses. Both cannot be fully right; the model sided with the second by treating the gradient as confounding-driven — a live, unsettled call shipped as near-elimination of H-22.