Pools three Harvard cohorts free of T2D/CVD/cancer at baseline: Nurses’ Health Study (NHS, 1980-2012), NHS II (1991-2017), and Health Professionals Follow-up Study (HPFS, 1986-2016) - combined n=213,799, 5,529,959 person-years, 20,514 incident T2D cases. Pooled multivariable HR for 1 egg/day increment = 1.14 (95% CI 1.07-1.20), i.e. 14% higher T2D risk, driven mainly by NHS (+19%) and NHS II (+15%); weaker in HPFS men (+7%). The paper’s own updated dose-response meta-analysis of 16 prospective cohorts (589,559 participants, 41,248 T2D cases) found pooled RR 1.07 (95% CI 0.99-1.15) per egg/day overall, but a strong region interaction (P-interaction <0.01): +18%/egg-day in US studies, null in Europe, -18%/egg-day in Asian studies. This is the modern, most-cited confirmation of the US/Europe/Asia gradient, and itself functions as a discovery hub for the region-stratified cohort list (overlaps heavily with Tamez 2016’s list). relevance_note: the Harvard-cohort primary explicitly named in the brief; its own meta-analysis is the cleanest single statement of the three-way US>Europe>Asia gradient.
Methodology
Prospective pooling of three Harvard cohorts free of T2D/CVD/cancer at baseline - Nurses’ Health Study (NHS, women, 1980-2012), NHS II (women, 1991-2017) and Health Professionals Follow-up Study (HPFS, men, 1986-2016), combined n=213,798, 5,529,959 person-years, 20,514 incident T2D cases. Egg intake from repeated FFQs entered as cumulative averages; Cox models (Model 3, fully adjusted) controlled for age, BMI, smoking, physical activity, alcohol, total energy and intake of foods co-consumed with eggs (bacon, red/processed meat, refined grains, potatoes, coffee, sugar-sweetened beverages) and overall diet quality (AHEI). Cohort HRs pooled by fixed-effects meta-analysis. The paper separately updates a dose-response meta-analysis to 16 prospective cohorts (589,559 participants, 41,248 cases), stratified by region.
Results
O-14 - Pooled Harvard 3-cohort egg-T2D HR 1.14 per additional egg per day
O-14 — Cox model adjusted for age, BMI, and diet/lifestyle covariates including intake of bacon, red/processed meat, refined grains, potatoes, coffee, sugar-sweetened beverages and diet quality; exposure from repeated (cumulative-average) FFQs. See
Link to originalMethodology
Prospective pooling of three Harvard cohorts free of T2D/CVD/cancer at baseline - Nurses’ Health Study (NHS, women, 1980-2012), NHS II (women, 1991-2017) and Health Professionals Follow-up Study (HPFS, men, 1986-2016), combined n=213,798, 5,529,959 person-years, 20,514 incident T2D cases. Egg intake from repeated FFQs entered as cumulative averages; Cox models (Model 3, fully adjusted) controlled for age, BMI, smoking, physical activity, alcohol, total energy and intake of foods co-consumed with eggs (bacon, red/processed meat, refined grains, potatoes, coffee, sugar-sweetened beverages) and overall diet quality (AHEI). Cohort HRs pooled by fixed-effects meta-analysis. The paper separately updates a dose-response meta-analysis to 16 prospective cohorts (589,559 participants, 41,248 cases), stratified by region.
Link to original
O-15 - Egg-T2D HR by Harvard cohort - positive in the two women cohorts, null in HPFS men
O-15 — Cohort-specific estimates before pooling; HPFS (men) crosses 1. See
Link to originalMethodology
Prospective pooling of three Harvard cohorts free of T2D/CVD/cancer at baseline - Nurses’ Health Study (NHS, women, 1980-2012), NHS II (women, 1991-2017) and Health Professionals Follow-up Study (HPFS, men, 1986-2016), combined n=213,798, 5,529,959 person-years, 20,514 incident T2D cases. Egg intake from repeated FFQs entered as cumulative averages; Cox models (Model 3, fully adjusted) controlled for age, BMI, smoking, physical activity, alcohol, total energy and intake of foods co-consumed with eggs (bacon, red/processed meat, refined grains, potatoes, coffee, sugar-sweetened beverages) and overall diet quality (AHEI). Cohort HRs pooled by fixed-effects meta-analysis. The paper separately updates a dose-response meta-analysis to 16 prospective cohorts (589,559 participants, 41,248 cases), stratified by region.
Link to original
O-16 - Updated 16-cohort meta-analysis - overall egg-T2D RR 1.07 per egg per day, non-significant
O-16 — Pooled random-effects estimate across the worldwide prospective-cohort literature (includes the three Harvard cohorts plus external US, European and Asian cohorts).
Link to original
O-17 - Meta-analysis region gradient - egg-T2D RR US 1.18, Europe 0.99, Asia 0.82, p-interaction 0.01
O-17 — Region stratification of the same pooled cohort set; the US stratum is dominated by the Harvard cohorts.
Link to original
O-18 - Harvard egg-T2D association stronger at lower overall diet quality (AHEI interaction)
O-18 — Effect-modification analysis stratifying the pooled association by Alternative Healthy Eating Index.
Link to original
O-19 - Substituting eggs with whole grains, nuts, yogurt or reduced-fat dairy lowers modelled T2D risk; other animal proteins no change
O-19 — Statistical substitution modelling (not observed behaviour): eggs rank worse than plant foods and dairy but equivalent to other animal-protein foods.
Link to original
Discussion
H-8 - Egg-T2D association is region-population-dependent - positive in US, null in Europe, null-to-inverse in Asia
H-8 — The paper’s central interpretive claim: the overall null masks a genuine, statistically-significant three-way regional difference rather than a single universal effect.
Link to original
H-9 - US egg-T2D association reflects residual confounding by the co-consumed Western dietary pattern, not a causal egg effect
H-9 — Authors conclude the higher US risk “may not reflect egg consumption per se but rather egg consumption habits”.
Link to original
A-5 - Region- and diet-quality-dependence plus incomplete attenuation point to the US egg-T2D signal being confounded by the Western dietary pattern
Reasoning (A-5): A true biological effect of eggs on glucose metabolism should be broadly portable across populations and should not depend on the background diet an egg is eaten with. The data show the opposite pattern on two independent axes. (1) By region: the per-egg/day RR is +18% in the US but null in Europe (0.99) and null-to-inverse in Asia (0.82), a statistically-significant interaction (p=0.01). Egg biochemistry does not change between continents; what changes is the food matrix eggs are eaten in - in the US eggs are typically co-consumed with bacon, red/processed meat, refined grains and sugary drinks (a diabetogenic Western pattern), whereas in Asia eggs enter prudent/traditional cuisines. (2) By diet quality within the US cohorts: the association is stronger at low AHEI (1.16) than high AHEI (1.09), p-interaction 0.03 - i.e. the “effect” grows precisely where the surrounding diet is worse, the signature of confounding by that diet rather than of the egg itself. The models already adjusted for bacon, red/processed meat, refined grains, potatoes and sugar-sweetened beverages, yet the association persisted; but the authors note adjustment for foods co-consumed with eggs “may not totally eliminate residual confounding”, and imperfectly-measured dietary-pattern covariates leave residual confounding that scales with how Western the diet is. Both interaction patterns are therefore jointly explained by egg intake acting as a marker of the Western dietary pattern, and are hard to reconcile with a portable causal effect of eggs. The inference is probabilistic (interactions can also arise from true effect-modification), so it raises but does not prove the confounding hypothesis.
Validity (step 6)
status: approved | reason_if_not_false: checked
Traced the step. The conclusion in the statement is comparative and hedged: the data are “more consistent with confounding by the Western dietary pattern … than with a universal causal effect.” Given the premises (significant region interaction p=0.01 with the gradient tracking how Western the food matrix is; significant within-US AHEI interaction p=0.03 with the association strongest where diet quality is worst; persistence after adjustment with authors flagging residual confounding), the step goes through on both halves of the comparison. (a) Against a universal/portable causal effect: any statistically significant effect-modification is by definition inconsistent with a universal (population-invariant) effect - this half is near-definitional given the premises. (b) For confounding-by-pattern specifically: confounding by a Western dietary pattern predicts exactly the observed directionality (strongest in the US and at low AHEI, i.e. where the pattern is most present), so the data raise its probability. The obvious undercutting defeater - a genuine biological effect that is truly modified by background diet - is explicitly acknowledged in the body and does not break the stated conclusion, because genuine effect-modification is itself not “a universal causal effect”, so it falls on the same side of the comparison the statement draws. The statement claims only “more consistent with”/“point to”, not proof, so no over-reach requiring correction. Holds as stated.
Link to original
A-6 - Substitution pattern - eggs worse than plant foods and dairy but equivalent to other animal proteins - fits eggs as an animal-food marker rather than a unique cause
Reasoning (A-6): Substitution modelling holds total energy fixed and asks what happens when one food displaces another, so the sign of each contrast reveals eggs’ rank relative to the substitute rather than an absolute effect. Two facts constrain the interpretation. (a) Replacing eggs with plant foods (whole grains, nuts) or dairy (yogurt, reduced-fat milk) lowers modelled risk by ~9-19% - but this is expected because those foods are themselves inversely associated with T2D, so a lower-risk result follows from the substitute’s benefit, not necessarily from eggs’ harm. (b) Replacing eggs with red/processed meat, poultry, fish or unprocessed meat produces no significant change - i.e. within the animal-food class, eggs are interchangeable with other animal proteins for T2D risk. If eggs carried a specific diabetogenic property (e.g. via cholesterol) beyond being an animal food eaten in a Western pattern, swapping them for other animal proteins should still move risk; it does not. The pattern therefore locates eggs as one member of an animal-food/Western-pattern cluster rather than a distinctive cause, supporting the confounding-by-pattern reading. Caveat: substitution estimates are model-derived, not observed behaviour, and inherit the same residual confounding as the primary analysis.
Validity (step 6)
status: corrected | reason_if_not_false: checked
Traced the step. The load-bearing move is: an egg→other-animal-protein substitution shows “no significant change” ⇒ eggs are no more diabetogenic than the animal-food class ⇒ eggs are a marker rather than a distinctive cause. Two links, of which the first has an undercutting defeater. “No significant change” is a failure to reject the null, not evidence of equivalence: a non-significant substitution contrast is equally produced by a genuinely underpowered comparison as by true equivalence, so “does not ⇒ eggs are no more diabetogenic (implies)” over-reads. Absence of evidence of a difference is not evidence of no difference. Hence the deductive “implies” fails; the weaker evidential form immune to the defeater - the null contrast is consistent with, and mildly supportive of, eggs-as-animal-food-marker - does hold, since equivalence remains one live reading of a null point estimate and it coheres with the confounding hypothesis. Note also the plant/dairy arm is correctly self-neutralised in the body (the risk drop there is attributed to the substitute’s own inverse association, so it carries no weight for eggs’ harm), leaving the whole inference resting on the animal-protein null. Corrected: “implies eggs are no more diabetogenic” → “is consistent with eggs being no more diabetogenic”; the downstream “supports the broader-pattern reading” is retained as evidential, not probative. A second, non-fatal caveat (already in the body) is that substitution estimates are model-derived and inherit the primary analysis’s residual confounding - this bears on premise truth (priced later), not on the corrected step.
Original
statement: “That swapping eggs for whole grains, nuts or dairy lowers modelled T2D risk while swapping eggs for other animal proteins (meat, poultry, fish) does not implies eggs are no more diabetogenic than the animal-food class they belong to, consistent with the association being driven by the broader dietary pattern rather than something specific to eggs.”
Link to original