O-108 - China-PAR U-shaped egg-CVD association lowest at 3 to under 6 eggs per week higher at both extremes
Reference (lowest-risk) category 3-<6 eggs/week. Cohort-stratified Cox regression, multivariable-adjusted. Intermediate category HRs (1-<3/week, 6-<10/week) were not accessible (paywalled full text); the reported extremes plus the interior reference establish the U-shape and its thresholds.
Link to original
Why this is evidence
Within a Chinese cohort the egg-CVD association is U-shaped (higher risk below ~1/week and above ~10/week, lowest at 3-<6/week), i.e. one universal non-linear dose-response. This lets H-50’s differential-exposure-range explanation reframe the “regional modification” as different regions sampling different arms of one curve (China Kadoorie sits entirely on the low, descending limb). H-22 (a true regional difference in the causal effect) and H-7 (monotone healthy-user confounding, which cannot generate harm at high intake) predict this pattern less well.
Likelihood
# E-127 (HC-2) — lone edge, one observation O-108 "China-PAR: within one Chinese cohort the egg-CVD
# association is U-shaped — lowest at 3-<6 eggs/wk, <1/wk HR 1.22, >=10/wk HR 1.39." Members in
# HC-2.hypotheses order [H-7, H-22, H-50]. Anchored on H-50 = 1 (best fit); others as ratios (rule 7).
lik_o108_H50 = 1.0 # anchor: a within-cohort U-shape is the direct signature of H-50's differential-
# exposure-range explanation — one universal non-linear dose-response, with regions
# (Kadoorie on the low descending limb) sampling different arms; A-41 (approved) confirms
# no monotone model fits this population
lik_o108_H22 = 0.4 # 0.4×: a within-Asia U-shape (harm at both extremes) undercuts a clean causal "inverse
# in Asia"; genuine region-modification predicts this pattern less well. Not lower
# because modification could in principle co-exist with a dose-shape
lik_o108_H7 = 0.35 # 0.35×: healthy-user/residual confounding is monotone and CANNOT generate the high-
# intake harm arm (HR 1.39 at >=10/wk); A-42 (approved) shows it DOES explain the
# inflated low-intake arm (HR 1.22, reverse-causation / low-SES), so partial fit — but
# the unexplained right arm makes it the poorest of the three
t_o108 = 0.62 # cap = trust_score of O-108's source S-81 = 0.68; docked to 0.62 for a specific
# weakness — the intermediate category HRs were paywalled/unread, so the "U" shape is
# inferred from the two reported extremes plus the interior reference (rule 4)
evidence("HC-2", ["O-108"], [lik_o108_H7, lik_o108_H22, lik_o108_H50], t=t_o108)