Primary data: Cohort of Swedish Men (COSM), 39,610 men aged 45-79, followed 1998-2012 (15y) for incident T2D via linked registers; 4,173 cases. Egg intake by FFQ. Compared to <1 egg/week, multivariable HRs were 0.98 (1-2/week), 1.11 (3-4/week), 1.11 (>=5/week), p-trend=0.06 - i.e. a weak, non-significant upward trend, functionally null. The paper also updates the dose-response meta-analysis to 12 prospective cohorts and reports the same US-vs-non-US split later reproduced by Drouin-Chartier 2020: HR per 3 eggs/week was 1.18 (95% CI 1.13-1.24) in 5 US studies (I2=0%) vs 0.97 (95% CI 0.90-1.05) in 7 non-US studies (mix of European + 1 Japanese). Functions as both a European primary (null-ish COSM result) and a second independent statement of the US/non-US heterogeneity. relevance_note: large null-to-weak-positive European primary; its meta-analysis is the second independent confirmation (after Tamez 2016) of the US vs non-US split.
Methodology
Cohort of Swedish Men (COSM): 39,610 men aged 45-79, single baseline FFQ egg exposure, followed 1998-2012 (15y) for incident T2D captured via linked national registers (4,173 cases). Cox models adjusted for age, BMI, physical activity, education, smoking, total energy, alcohol, coffee, red/processed meat, fish, fruit, vegetables, white bread, fibre and CVD history; a further model added dietary cholesterol and protein. The paper also updates a dose-response meta-analysis to 12 prospective cohorts (287,963 participants, 16,264 cases), stratified into US vs non-US studies and tested for non-linearity.
Results
O-29 - Cohort of Swedish Men - egg intake and T2D functionally null (p-trend 0.06) with register-based outcomes
O-29 — Adjusted for age, BMI, physical activity, education, smoking, energy, alcohol, coffee, red/processed meat, fish, fruit, vegetables, white bread, fibre and CVD history; T2D captured via linked national registers, reducing outcome misclassification. See
Link to originalMethodology
Cohort of Swedish Men (COSM): 39,610 men aged 45-79, single baseline FFQ egg exposure, followed 1998-2012 (15y) for incident T2D captured via linked national registers (4,173 cases). Cox models adjusted for age, BMI, physical activity, education, smoking, total energy, alcohol, coffee, red/processed meat, fish, fruit, vegetables, white bread, fibre and CVD history; a further model added dietary cholesterol and protein. The paper also updates a dose-response meta-analysis to 12 prospective cohorts (287,963 participants, 16,264 cases), stratified into US vs non-US studies and tested for non-linearity.
Link to original
O-30 - Wallin 12-cohort meta-analysis - overall egg-T2D HR 1.03 per 3 servings per week, non-significant, no nonlinearity
O-30 — Random-effects pooling of the prospective-cohort literature; the flat linear dose-response argues against a threshold effect within observed intake ranges.
Link to original
O-31 - Wallin meta-analysis US-vs-non-US split - egg-T2D HR 1.18 (US, I2 0%) vs 0.97 (non-US)
O-31 — A second, independent restatement (before Drouin-Chartier 2020) of the US/non-US heterogeneity, but built on the same overlapping cohort set.
Link to original
Discussion
H-13 - Egg consumption is a marker of the Western dietary pattern that drives T2D, not a causal factor
H-13 — Authors’ preferred explanation for why the association appears only where eggs are embedded in a Western diet.
Link to original
H-14 - Egg-T2D association is positive in US populations but null in non-US (European and Japanese) populations
H-14 — The heterogeneity is the paper’s headline: a single pooled estimate is misleading because the US and non-US strata point different ways.
Link to original
A-10 - US-only egg-T2D signal plus null after cholesterol-protein adjustment supports eggs marking a dietary pattern rather than causing T2D
Reasoning (A-10): Two features of the data jointly discriminate a causal from a marker interpretation. First, the candidate causal mediators of an egg effect are its cholesterol and protein content, which are constant across countries; yet in COSM, additionally adjusting the model for cholesterol and protein intake did not substantially alter the (already near-null) egg-T2D estimate. If cholesterol or protein were driving risk, controlling the association’s own proposed mechanism should have moved it - it did not, arguing the residual signal is not nutrient-mediated. Second, the association is present only in US cohorts and, notably, is internally consistent there (I2 0% at HR 1.18) while absent in non-US cohorts (0.97). A universal biological effect of eggs cannot switch on only in the US; what differs is the diet eggs accompany - a Western pattern (red/processed meat, fried potatoes, high-fat dairy, sweets, refined grains) in the US versus prudent/traditional patterns where eggs are eaten in Japan. Register-based outcome ascertainment in COSM (low misclassification) makes it unlikely the non-US null is an artefact of missed cases. Taken together, egg intake behaving as a proxy for the local dietary pattern explains both the geographic on/off pattern and the mechanism-adjustment null more parsimoniously than a direct causal effect. The argument is inferential, not deductive: unmeasured effect-modification by a genuinely causal egg effect remains logically possible.
Validity (step 6)
status: approved, reason_if_not_false: checked.
Traced both prongs. Prong 1 (mediator-adjustment): if eggs raise T2D risk through their cholesterol/protein content, conditioning on cholesterol and protein intake should attenuate the egg coefficient; it did not move → those specific nutrients are not the operative mediators. The step is a valid mediation inference. It is weak here (COSM is already near-null, so there is little to attenuate and the mediator test was run in a cohort showing no effect), but weakness is a matter of strength, not validity, and the statement is framed as “better explained by,” not a proof. Prong 2 (geographic on/off): a universal biological mechanism cannot switch on only in US cohorts, so an I2=0% signal at 1.18 confined to the US while non-US is 0.97 points to something that co-varies with geography — the accompanying diet — rather than the egg itself.
The statement is a comparative/parsimony claim (“better explained by … than by”), and both prongs push the same way, so the evidential step holds. The obvious undercutting defeater — effect-modification of a genuinely causal egg effect by the surrounding Western diet, which would also produce a US-only signal — is explicitly acknowledged and only downgrades the argument to evidential (which it already is), not refuted. Approved as stated.
Link to original