EpiStack

egg-diabetes-observational

8 min read

Orientation — egg intake and type-2 diabetes, observational (searcher 1b)

Slice: observational egg-intake evidence on type-2 diabetes, split into (i) egg→incident-T2D in prospective cohorts, where a US-vs-Asia regional heterogeneity replicates across independent meta-analyses, and (ii) dedicated analyses of the diabetic subgroup specifically, where egg intake associates with worse CVD/mortality outcomes in people who already have diabetes. 10 source nodes minted (budget: 10 write / ~20 read), spread across distinct cohort families per the brief rather than piling onto any one region. ~20 primaries/reviews read at varying depth; several discovery-hub meta-analyses skimmed far more cheaply and harvested for reference lists.

Topic: US cohorts — egg→T2D risk elevated

  1. S-4 - Drouin-Chartier 2020 AJCN - egg consumption and T2D in Harvard NHS-NHS II-HPFS pooling + updated meta-analysis.md — Harvard NHS/NHS II/HPFS pooled (n=213,799, 20,514 T2D cases): +14%/egg-day pooled. Also carries its own 16-cohort updated meta-analysis with a clean 3-way region-interaction test (US +18%/Asia −18%/Europe null) — the single best-documented statement of the whole regional-heterogeneity finding, and itself a mini discovery-hub.
  2. S-1 - Djoussé 2009 Diabetes Care - egg consumption and T2D in Physicians’ Health Study and Women’s Health Study.md — the founding primary (Physicians’ Health Study men + Women’s Health Study women in one paper), most-cited single US result (HR 1.58 men / 1.77 women, ≥7 eggs/wk vs none). Explicitly named in brief.
  3. S-6 - Djoussé 2016 Clinical Nutrition - egg consumption and T2D among African Americans, Jackson Heart Study.md — a demographically distinct US cohort (African American, Jackson Heart Study) by the same lead author. Useful complication: positive for prevalent diabetes but null for incident diabetes (HR 1.17, CI 0.81–1.70) — shows the US-elevated pattern is not perfectly uniform even within the US once you require a genuinely prospective (not cross-sectional/prevalent) endpoint.

Topic: European cohorts — null/weak

  1. S-8 - Wallin 2016 Diabetologia - egg consumption and T2D in the Cohort of Swedish Men + updated dose-response meta-analysis.md — Cohort of Swedish Men (n=39,610): functionally null trend (HR 0.98–1.11, p-trend=0.06). Also carries its own 12-cohort updated meta-analysis independently reproducing the US (+18%/3-per-week) vs non-US (null, −3%) split — a second, earlier confirmation of the same heterogeneity Drouin-Chartier later restates.

Topic: Asian cohorts — null/inverse, with one clear counter-example

  1. S-9 - Kurotani 2014 BJN - cholesterol and egg intakes and T2D risk, Japan Public Health Center-based Prospective Study.md — JPHC, Japan: 40% lower T2D risk in men eating 2–<4 eggs/week vs <1/week. Large, well-established Japanese national cohort.
  2. S-11 - Lee & Kim 2018 Nutrition Research and Practice - egg consumption and lower T2D risk in Korean men, KoGES.md — KoGES, Korea: also 40% lower risk in men (HR 0.60, 95% CI 0.37–0.97); null in women (wide CI).
  3. S-13 - Wang, Li and Shi 2020 BJN - higher egg consumption and increased diabetes risk, China Health and Nutrition Survey.md — China Health and Nutrition Survey: the counter-example. A genuinely longitudinal Chinese cohort (7 survey waves, 1991–2009) finding elevated risk (ORs 1.25–1.37 above the lowest quartile). Shows the US/Asia split is a real central tendency, not a clean geographic partition — China itself is not uniformly null/inverse the way Japan and Korea are here.

Topic: diabetic-subgroup CVD/mortality interaction (thread ii)

  1. S-14 - Jang 2018 Nutrition and Diabetes - egg consumption and CVD risk interacting with T2DM status, KoGES Ansung-Ansan.md — KoGES Ansung-Ansan, Korea. Paper’s designed focus (not a post-hoc subgroup cut): among diabetics, highest vs lowest egg intake gave HR 2.81 (95% CI 1.25–6.30) for incident CVD; null in non-diabetics (HR 1.03). Notable for being an Asian cohort — same population whose general T2D-incidence association is null/inverse (cf. topic above), yet shows a large adverse CVD signal once restricted to existing diabetics.
  2. S-16 - Trichopoulou 2006 J Intern Med - diet, physical activity and mortality among diabetic adults in a Greek population cohort.md — a genuine diabetics-only cohort (1,013 diabetics inside Greek EPIC, not a subgroup cut): adverse mortality signal, CVD-mortality HR 1.54 (95% CI 1.20–1.97) per 10 g/d higher egg intake. European, complementing Jang’s Asian result.
  3. S-17 - Díez-Espino 2017 Clinical Nutrition - egg consumption and CVD by diabetic status, PREDIMED.md — PREDIMED (Spain), a dedicated diabetic-status interaction test that found no interaction (diabetics HR 1.33, 95% CI 0.72–2.46; non-diabetics HR 0.96, 95% CI 0.33–2.76 — both null, wide, overlapping). The tempering counter-example to 8 and 9: the “worse for diabetics” signal is not universal across the three dedicated interaction/diabetics-only analyses found.

search_scope

Started from the region-stratified dose-response meta-analyses named in the brief and read their full text/summaries for included-cohort lists with region labels: Tamez et al. 2016 (BJN; 10 cohorts, US RR 1.47 vs null elsewhere — this yielded most of the US/Europe/Finland/Japan cohort roster directly), Wallin et al. 2016 (Diabetologia; itself also the COSM primary, 12-cohort update, US 1.18 vs non-US 0.97 per 3/wk), Drouin-Chartier et al. 2020 (AJCN; itself also the NHS/HPFS primary, 16-cohort update with explicit 3-way US/Europe/Asia region-interaction test), and Shin et al. 2013 (AJCN CVD+diabetes+diabetic-subgroup meta — used mainly for confirmation that “high egg intake ~ elevated CVD risk in the T2D subpopulation” is a described meta-analytic finding, not fully mined for its unique cohort list beyond overlap with the other three). Also attempted the Dove Medical Press (DMSO.S58668) systematic review specifically on egg-CVD-in-diabetics as a thread-ii discovery hub; full text was paywalled (403) but its cited summary statistics (“7,549 diabetic patients across 4 cohorts”; “606 deaths among 1,529 diabetics across 3 cohorts”) led directly to identifying Trichopoulou 2006 by name search. Beyond the hubs: targeted keyword searches per named cohort family (Djoussé + Physicians’ Health Study / Women’s Health Study / Cardiovascular Health Study / Jackson Heart Study; EPIC-InterAct + eggs + diabetes; China Kadoorie Biobank + eggs + diabetes; Shanghai / Singapore Chinese Health Study + eggs; Jiangsu / Hangzhou / China Health and Nutrition Survey + eggs + diabetes; Korea + eggs + diabetes, twice over — KoGES general-population and KoGES Ansung-Ansan diabetic-interaction; Tehran Lipid and Glucose Study + eggs + diabetes; PREDIMED + eggs + diabetic status; Trichopoulou + diabetic + eggs + mortality). Citation counts pulled via the Semantic Scholar API (through WebFetch, not WebSearch) — got exact counts for 2 of 10 before hitting a 429 rate limit; left unknown on the rest per spec.

exclusions

  1. Radzevičienė & Ostrauskas 2012, Public Health Nutrition (Lithuania, 234 T2D cases/468 controls) — case-control design, not a prospective cohort; out of scope for the “incident T2D in prospective cohorts” thread.
  2. Ni et al. 2020, Medicine (Baltimore) (Hangzhou, China, n=3,298) — cross-sectional observational study (no incidence/follow-up), not a cohort.
  3. Shi et al. 2011, Nutrition (Jiangsu, China, n=2,849) — repeated cross-sectional household-survey design, not prospective; found a positive OR in women (1.66–3.01) directionally consistent with the CHNS finding (S-13) but doesn’t meet the cohort-design bar, so not minted.
  4. Shin et al. 2013 (AJCN) and Li et al. 2013 (Atherosclerosis, sometimes the paper actually meant by a loose “Wallin 2013” citation) — both discovery-hub meta-analyses per the brief’s instruction to mine, not node, review/meta articles; used for corroboration and cohort-roster cross-checking, not minted.

Gaps / not found within budget

  1. No standalone EPIC-InterAct primary analysis of egg intake and incident T2D was found — egg only appears inside broader multi-food substitution/protein-source papers (e.g. meat-substitution, dietary-protein-sources analyses) without an isolatable egg-only HR. The brief’s expectation of a European “~null” EPIC-InterAct primary is therefore not represented by its own node; Wallin 2016 (COSM, Sweden) and the null components of Tamez/Drouin-Chartier’s meta-analyses are the closest European coverage in this pool.
  2. No Tehran Lipid and Glucose Study (or other Iranian cohort) egg-T2D primary was found — searches surfaced only papers treating eggs as one of several fat/lipid sources, not an egg-specific incident-T2D analysis.
  3. China Kadoorie Biobank itself has no dedicated egg-T2D paper (its diabetes-diet papers cover red meat/poultry/fish, not eggs); China thread-(i) coverage rests on the China Health and Nutrition Survey (S-13) alone once the two cross-sectional Chinese studies are excluded (#2–3 above).
  4. Women’s Health Initiative (AJCN 2022, “Eggs, dietary cholesterol, choline, betaine, and diabetes risk”) — a plausible fourth distinct US cohort family, located but not read (403-blocked, and general web-search budget ran out before a workaround was tried); flagged for a future searcher/step-2 pass rather than minted on secondhand snippets alone.
  5. Djoussé et al. 2010, Cardiovascular Health Study (elderly US cohort) — read via search snippet only (HR 1.81, 95% CI 0.77–4.22, i.e. null/underpowered); not minted purely for the 10-source budget ceiling, not a quality judgment — a natural 11th pick if step 2 wants a 4th US cohort or an elderly-population data point.

Graph View

Show model graphinteractive Bayesian view — separate from the graph above
added after the FLF submission deadline