EpiStack

00-consolidated-pool

9 min read

Step 1 — Consolidated source pool (consolidator 1c)

Main question: “Is habitual egg consumption net beneficial, harmful, or neutral for human health — overall, for whom, and at what level of intake?”

This is the merged overview step 2 (curate) starts from. It lists all 20 source nodes (S-1…S-20) grouped by line of evidence, best-first within group; records the two cross-slice shared-cohort overlaps step 2’s data-basis work must handle; unions the three searchers’ exclusions; and closes with a short coverage read. The pool came from three blind-parallel slices (1a’s plan: 00-search-plan.md; searchers: mechanism-surrogate.md, cohorts-hard-endpoints.md, heterogeneity-subgroup-dose.md).

Pool description

20 primaries spanning 1965–2020, balanced across the debate’s three FLF-seeded design types (which are also the intended downstream cluster boundaries — surrogate and hard endpoints must not be merged): 11 surrogate-endpoint (blood-lipid) studies and 9 hard-endpoint (CVD-event / mortality / incident-diabetes) studies. Finding-direction is balanced within every group (see below). Motivatedness is concentrated and patterned: egg-industry funding (Egg Nutrition Center / American / Australian Egg Board, largely Fernandez-lab) is confirmed or inferred on 5 sources — S-1, S-2, S-15, S-17, S-20 — and, notably, sits almost entirely on the null/pro-egg surrogate studies (see coverage note 2). Several “egg” exposures are actually dietary-cholesterol proxies (S-3, S-6, S-8), era-appropriate but worth tagging when step 3 attributes observations.

Sources by line of evidence

1. Mechanism / surrogate endpoint — dietary cholesterol & whole egg → blood lipids

Controlled-feeding / metabolic-ward studies and short RCTs in general/healthy adults. Endpoint = surrogate throughout. Split 3 rise / 4 null-or-favorable.

Independent controlled-feeding — LDL rises (with diminishing returns / homeostatic compensation):

  1. S-10 - A dose-response study of the effects of dietary cholesterol on fasting and postprandial lipoproteins in healthy young men — modern 4-way within-subject crossover (0/1/2/4 yolks/day), LDL-C rises then plateaus; best single dose-response primary in the pool.
  2. S-8 - Serum cholesterol response to changes in the diet - the effect of dietary cholesterol (metabolic-ward dose-response series) — foundational 1965 Keys metabolic-ward series behind the cholesterol-response equation; total serum cholesterol (pre-LDL-fractionation).
  3. S-12 - Ingestion of egg raises plasma low density lipoproteins in free-living subjects — ecological single-egg-a-day add-on in lacto-vegetarians; LDL +12%, ApoB +9%, HDL unchanged. (Plan mis-seeded this as “Sacks NEJM, null anchor”; it is Lancet 1984 and a rise finding — corrected by the searcher.)

Whole-egg RCTs — null / favorable LDL (HDL up, particle-quality, or poor absorption): 4. S-17 - Intake of up to 3 eggs per day is associated with changes in HDL function and increased plasma antioxidants in healthy young adults — 0–3 eggs/day; larger LDL/HDL particles, HDL antioxidant function, no LDL-C/ApoB rise. (Fernandez lab; ENC funding inferred.) 5. S-15 - Dietary cholesterol from eggs increases plasma HDL cholesterol in overweight men consuming a carbohydrate-restricted diet — 3 eggs/day on low-carb background raises HDL, LDL/HDL ratio preserved. (ENC-funded, Fernandez lab.) 6. S-20 - Daily egg consumption in hyperlipidemic adults - effects on endothelial function and cardiovascular risk — higher-risk (hyperlipidemic) population, 2 eggs/day vs a real cholesterol-free substitute; LDL essentially unchanged. (ENC-funded + CDC grant.) 7. S-19 - Dietary cholesterol contained in whole eggs is not well absorbed and does not acutely affect plasma total cholesterol concentration — absorption/bioavailability mechanism: intact-egg cholesterol poorly absorbed, distinguishing whole-egg from purified-cholesterol feeding. (Kim & Campbell / Purdue; ENC funding inferred.)

2. Observational cohorts — hard endpoints (CVD events + mortality), general populations

One primary per distinct cohort family (7 families, 4 regions). Endpoint = hard. Split 2 harm / 2 null / 3 protective-or-nonlinear.

Harm-leaning (higher intake → higher risk):

  1. S-7 - Zhong 2019 JAMA - Dietary cholesterol and egg intake linked to higher incident CVD and mortality in 6 pooled US cohorts — 6-cohort NHLBI Lifetime Risk Pooling Project (ARIC/CARDIA/Framingham/Jackson/MESA), n=29,615; +6% CVD, +8% mortality per half-egg/day. Headline anti-egg anchor; 225 cites.
  2. S-14 - Djoussé 2008 Physicians Health Study - Egg consumption linked to higher mortality, especially in diabetics — 21,327 male US physicians; null on MI/stroke, mortality rises with intake, ~2× in the diabetic subgroup.

Null (no significant association): 3. S-9 - Drouin-Chartier 2020 BMJ - Egg intake not associated with CVD risk in NHS-NHSII-HPFS, updated meta-analysis — Harvard NHS/NHSII/HPFS, n=215,618; ≥1 egg/day not associated with CVD. Direct counterweight to Zhong. (Its meta-analysis section is a discovery hub, not separate data.) 4. S-11 - Dehghan 2020 PURE - Egg intake not associated with blood lipids, CVD, or mortality across 50 countries — PURE (146,011 / 21 countries) + ONTARGET/TRANSCEND (31,544); null on lipids, mortality, CVD. Most geographically diverse.

Protective or mixed/nonlinear: 5. S-13 - Qin 2018 Heart - Daily egg consumption associated with lower CVD risk in China Kadoorie Biobank — 0.5M Chinese adults (largest single cohort); daily eggs 11–26% lower risk across endpoints, clean dose-response. Central protective anchor. 6. S-16 - Key 2019 Circulation - Egg intake modestly inversely associated with ischemic heart disease in EPIC — pan-European EPIC, n=409,885; eggs HR 0.93/20g/day but nonsignificant after a reverse-causation check — fragile. 7. S-18 - Nakamura 2004 NIPPON DATA80 - Egg consumption and a U-shaped mortality relationship in a Japanese national cohort — Japanese national cohort (n≈9,263); U-shaped, both very-low and very-high intake worse than ~1/day. Nonlinearity flag.

3. Heterogeneity — for whom & at what level (subgroup / dose / responder)

Endpoint mixed — tagged per source. Split pro/neutral vs anti holds within each sub-topic.

Type-2 diabetes (RCTs + dedicated diabetic cohort):

  1. S-1 - DIABEGG high-egg diet and cardiovascular risk factors in type 2 diabetes, 3-month RCT — RCT n=140, 2 eggs/day vs <2/week, no adverse lipid change. Pro-egg. Surrogate. (Australian Egg Corporation-funded.)
  2. S-3 - Dietary fat and cholesterol and CVD risk among women with type 2 diabetes, Nurses’ Health Study — dedicated T2D-women cohort (n=5,672); dietary cholesterol → CVD (RR 1.37 per 200mg/1000kcal). Anti-egg. Hard. Independent (NIH). Also the slice’s hard-endpoint dose-response source.
  3. S-2 - One egg per day vs oatmeal breakfast — inflammation and cardiometabolic markers in type 2 diabetes RCT — crossover RCT n=29 T2D; egg period lower TNF-α/AST, no adverse change. Pro/neutral. Surrogate. (Heaviest COI: ENC/Fernandez.)
    • cross-ref: S-14 ...’s diabetic subgroup (mortality ~2× in diabetics) lives in group 2 but bears here.

Dose-response above ~1 egg/day: 4. S-4 - Egg consumption and risk of type 2 diabetes in men and women, dose-response in Physicians’ and Women’s Health Studies — PHS+WHS, graded incident-T2D risk (≥1/day RR 1.58 men / 1.77 women). Anti-egg. Hard. Independent. Clearest explicit dose-response primary.

  • cross-refs (dose-graded findings housed elsewhere): S-10 ... 0/1/2/4 yolks, S-8 ... 50–1450 mg, S-7 ... per half-egg/day, S-13 ... dose-response trend, S-17 ... 0–3 eggs/day.

Inter-individual variation — responder / genotype: 5. S-5 - Characteristics of human hypo- and hyperresponders to dietary cholesterol — stable, lifestyle-unpredictable hypo-/hyperresponder classification. Basis for “population mean overstates harm for hypo-responders.” Surrogate. 6. S-6 - Apolipoprotein E polymorphism and serum lipid response to dietary fat and dietary cholesterol — ApoE 4/4 carriers ~10% total-cholesterol rise vs minimal in 3/3. Anti-egg-for-ApoE4. Surrogate. (Manipulates dietary cholesterol generally, not egg-branded.)

Cross-slice data-independence notes (for step 2’s data-basis / D-node work)

Slices 2 and 3 worked blind and each drew from the same US cohort families, so two source pairs sit on overlapping raw datasets. These are not duplicates (different exposure and/or endpoint — both kept, no duplicate_of set), but step 2 should give each shared dataset a common data-basis (D) node so steps 4/5 don’t treat them as independent evidence:

  1. {S-3 ..., S-9 ...} — both built on the Nurses’ Health Study. S-3 = T2D-women subset, exposure dietary cholesterol, 1980–1998; S-9 = general NHS/NHSII/HPFS, exposure egg intake, to 2012+. Different exposure + population; neither covers the other.
  2. {S-4 ..., S-14 ...} — both use Physicians’ Health Study (men). S-14 = PHS-men, outcome CVD/mortality; S-4 = PHS-men + WHS-women, outcome incident T2D. Different endpoint; S-4 adds an independent women’s cohort. Neither covers the other.

(Also worth D-node attention as a shared construct, not dataset: S-3, S-6, S-8 use dietary cholesterol as the egg proxy rather than eggs directly.)

Union of exclusions (all three searchers)

Non-primary / discovery-hub (correctly not noded):

  1. Qin response/commentary letter, PMID 30309864 — letter on S-13, not primary. (slice 2)
  2. Weggemans/Zock/Ordovas/Katan 2001 (Atherosclerosis) ApoE pooled analysis, n=395 — review-like hub. (slice 3)

Out of scope for the slice that found it (left for the correct slice, per the cross-slice rule): 3. Kim JE et al., egg CVD-risk-factors in prediabetes/T2D — diabetic population → slice 3. (found by slice 1; a different Kim paper from S-19) 4. Djoussé/PHS general-population CVD-mortality/heart-failure egg papers → slice 2. (slice 3) 5. Hu 1999 (NHS/HPFS) and Zhong 2019 — general-population, reserved for slice 2. (slice 3) 6. JACC dairy/animal-food dietary-pattern paper — doesn’t isolate egg exposure. (slice 2)

Redundant with a minted node (same trial/cohort family, dropped for budget): 7. Fuller 2018 AJCN — DIABEGG weight-loss/follow-up phase, same trial as S-1. (slice 3) 8. Nakamura 2018 (Japanese women re-eval) & Nakamura 2006 BJN (JPHC CHD) — redundant with NIPPON DATA80 / S-18. (slice 2)

In-scope real primaries cut for budget only (candidates for a top-up round): 9. Harman/Leeds/Griffin 2008 (Eur J Nutr) — parallel-group, energy-restricted, 2 eggs/day, LDL does not rise; no egg-industry funding indicated → would supply the missing independent whole-egg-null cell (see coverage note 2). (slice 1) 10. Herron/Fernandez hypo-/hyperresponder egg-feeding studies (e.g. Herron 2003 J Nutr) — egg-specific responder primaries the responder axis currently lacks. (slice 3) 11. Guangzhou Biobank Cohort (Zhang/Xu, Eur J Nutr 2019, n=28,024, null) — a distinct second Chinese cohort. (slice 2) 12. Ginsberg 1995 ATVB (healthy young women dose-response); Greene 2006 Nutr Metab (Fernandez, elderly, particle characteristics). (slice 1) 13. SUN Project (Zazpe 2011), Korea (KoGES/KNHANES), Rotterdam — optional cohort families, not pursued. (slice 2)

Structural gap (fit no slice as scoped): 14. Dawber 1982 (Framingham, cross-sectional egg vs serum cholesterol) — surrogate outcome but observational design; neither slice 1 (feeding/RCT) nor slice 2 (hard endpoint) claims it. (slice 1)

Coverage read (feeds the audit; detail returned to orchestrator)

  1. Design types — both well covered (11 surrogate / 9 hard); kept in separate groups. Adequate.
  2. Finding-direction × funding confound (surrogate slice) — the 3 independent controlled-feeding studies (S-10, S-8, S-12) all find an LDL rise; all 4 null/favorable whole-egg RCTs (S-17, S-15, S-20, S-19) carry egg-industry funding (confirmed/inferred). The independent-and-null cell is empty. Not a balance hole in count, but a confound step 4 should not miss; top-up candidate = Harman 2008 (#9).
  3. Distinct cohorts — 7 families (NHLBI-pooled, Harvard, PURE, Kadoorie, PHS/WHS, EPIC, NIPPON DATA80) ≥ plan’s ≥6. Adequate. Named-but-missing: UK Biobank (searched, no egg-isolating hard-endpoint primary found).
  4. Responder / genotype axis — thinnest: 2 sources (S-5, S-6), and S-6 is a generic-dietary-cholesterol (non-egg) ApoE study. Thin; top-up = Herron/Fernandez egg-specific responder primary (#10).

Graph View