The vegetarian/vegan pattern leg, anchored on two sources: Orlich 2013 (the Adventist Health Study
2 [AHS-2] prospective cohort — 73,308 analytic Seventh-day Adventists, 2,570 deaths, mean follow-up
5.79 y, Cox regression) for the single-cohort detail, and Landry 2024 (a gold-tier umbrella of 21 SRs,
AMSTAR-2 + GRADE) for the pooled multi-cohort CV picture. The certainty ceiling on both is confounding,
not sampling — the whole literature is observational, and the umbrella’s breadth does not buy
independence from that confound (it re-pools the same cohorts, Orlich’s AHS-2 among the family). Orbits
the dietary-patterns nucleus Mediterranean Diet and Cardiovascular Events.
The graded-pattern effect estimates (vs nonvegetarian reference)
Diet was set at baseline by a >200-item FFQ and cut into 5 patterns by frequency of animal-food intake (vegan / lacto-ovo / pesco / semi-veg / nonveg). Adjusted HRs, all vegetarians combined and by pattern:
| Contrast | All-cause HR (95% CI) | Note |
|---|---|---|
| All vegetarians combined | 0.88 (0.80-0.97) | headline; ~12% lower |
| — men | 0.82 (0.72-0.94) | significant |
| — women | 0.93 (0.82-1.05) | CI crosses 1 |
| Vegan (both sexes) | 0.85 (0.73-1.01) | CI crosses 1 |
| Lacto-ovo | 0.91 (0.82-1.00) | borderline |
| Pesco | 0.81 (0.69-0.94) | steepest, significant |
| Semi-vegetarian | 0.92 (0.75-1.13) | null |
| Vegan, men only | 0.72 (0.56-0.92) | significant |
| Pesco, men only | 0.73 (0.57-0.93) | significant |
The all-cause signal is modest (~10-20%) and graded but not monotone by “strictness” — pesco- and vegan-in-men are the steepest arms, semi-vegetarian is null; more animal-food restriction does not buy a larger reduction here. Absolute scale: the standardized mortality rate was 6.61/1000 person-years in nonvegetarians vs 5.40 (vegan) / 5.61 (lacto-ovo) / 5.33 (pesco) — a ~1/1000-person-year gap over a short horizon -> Baseline Risk and the Relative-Absolute Split.
Cause-specific: cardiovascular and a diabetes/renal channel; cancer null
- Cardiovascular: all-veg combined CVD mortality 0.87 (0.75-1.01); driven by men (CVD 0.71 [0.57-0.90]; IHD 0.71 [0.51-1.00]), null in women (CVD 0.99 [0.83-1.18]).
- Endocrine / renal (“other” mortality): the largest point estimates — renal mortality 0.48 (0.28-0.82) and endocrine mortality 0.61 (0.40-0.92) both sexes; sharper in men (renal 0.42 [0.19-0.91]; endocrine 0.48 [0.25-0.92]). The specific-cause subsets: renal-failure death 0.26 (0.12-0.57); diabetes death 0.53 (0.32-0.89), and 0.27 (0.11-0.66) in men. Consistent with the cohort’s prior vegetarian-diet associations with lower incident diabetes, hypertension, and metabolic syndrome — a plausible metabolic channel, but built on 67 renal and 104 endocrine deaths.
- Cancer: no significant reduction (0.92 [0.78-1.08] both sexes) — the authors flag heterogeneous cancer + insufficient power at early follow-up, not evidence of no effect (the expectancy test).
- Stroke: not reduced (1.10 [0.82-1.47] both sexes) — a genuine null arm, not a garnish.
- BMI sensitivity: adding BMI to the model moved the combined HR only to 0.90 (0.82-0.98) and «did not consistently move results toward the null» (Orlich et al., 2013) — so adiposity is not the whole story (or BMI is an over-adjustment mediator), but it does not vanish the signal either.
The sex interaction — a real effect-modification signal, mechanism unknown
(Orlich et al., 2013) «Associations in men were larger and more often significant than were those in women.» A formal test found significant diet×sex interaction for CVD mortality (P = .01), but not for all-cause or other categories — so the sex difference is statistically supported for the CV arm specifically, a descriptive pattern elsewhere. This replicates prior Adventist-cohort sex modification of the vegetarian-IHD association. The mechanism is unresolved: the authors note the nutrient profiles of male vs female vegetarians did not differ strikingly, so it may be a biological rather than a dietary-composition difference. This is a route-(b) effect-modification claim (positive interaction evidence, not a baseline-risk artifact), though from a single cohort.
Why certainty stays low — the healthy-adherent confound is the ceiling
This is the canonical observed-healthy-population case: an Adventist vegetarian is not a random person minus meat. Standardized baseline table — vegetarians (esp. vegans) were older, more educated, leaner (vegan BMI 24.1 vs nonveg 28.3), and drank/smoked far less (vegans 98.8% non-drinkers, 85% never-smokers; nonvegetarians 83.4% / 75.7%) and exercised more. The vegetarian “exposure” travels bundled with every other large mortality lever -> a pattern association is not evidence for any single component (the diet, the low BMI, the nuts, or the abstinence).
- The authors are explicit about the residual: «Observed mortality benefits may be affected by factors related to the conscious lifestyle choice of a vegetarian diet other than dietary components. Potential for uncontrolled confounding remains.» (Orlich et al., 2013)
- The adjustment set is unusually rich (age, race, region, income, education, marital status, smoking [8 levels], alcohol [5], exercise [5], sleep, energy, menopause/HRT — Table 1 footnote), and the cohort’s near-uniform low tobacco/alcohol means «the low use of tobacco and alcohol, making residual confounding from these unlikely» (Orlich et al., 2013). So this cohort reduces the usual smoking/alcohol confound better than most — but adjustment is the weak check; there is no MR instrument and no whole-diet RCT here, so the association is not adjudicated causal -> The U-Shaped Association Artifact (adjudication-route framing).
- Diet measured once, at baseline — «the analysis relies on a single measurement of diet at baseline» (Orlich et al., 2013); pattern misclassification over 5.79 y -> Measurement Error in Dietary Assessment.
- Short follow-up biases toward the null, not away: «Early follow-up analysis may thus have bias toward the null, and true associations may remain undetected.» (Orlich et al., 2013) So the modest HRs may understate a true effect — the confound and the attenuation push in opposite directions, and neither is quantified.
”Vegetarian” is not one exposure — the EPIC-Oxford divergence (transportability)
The prior British cohort (EPIC-Oxford, Key 2009) found no all-cause mortality advantage for vegetarians; AHS-2 does. Orlich’s own resolution is that the label names different real exposures: «It appears that British vegetarians and US Adventist vegetarians eat somewhat differently.» AHS-2 vegans ate far more fibre and vitamin C (mean fibre 45.6 g/d [men] / 47.3 [women] vs EPIC-Oxford 27.7 / 26.4; vitamin C 224/250 vs 125/143 mg/d) (Orlich et al., 2013). Two cohorts, same word, materially different diets and motivations (perceived-healthfulness vs ethical/environmental) — so a mortality effect attaches to a specific vegetarian diet in a specific population, and does not transport by the label alone. This sharpens the “does the food-category label do work over its components?” question -> Is the Food Category Doing Any Work, and means the pattern must be specified by its actual food/nutrient content, not its name -> The Comparator Problem (the nonvegetarian reference is itself a relatively healthy Adventist, which attenuates every contrast here).
The pooled multi-cohort picture — Landry 2024 umbrella (21 SRs)
(Landry et al., 2024) Landry pools the observational-cohort literature (general population, «presumably healthy adults» excluding diagnosed cardiometabolic disease; comparator = non-vegetarian/omnivorous diet) and grades each arm with GRADE:
| Outcome (vegetarian incl. vegan vs non-veg) | Pooled estimate | COE |
|---|---|---|
| CVD incidence | RR 0.85 (0.79, 0.92), I2=68% | moderate — highest arm |
| CHD incidence | RR 0.79 (0.71, 0.88), I2=67% | moderate |
| CVD mortality | HR 0.92 (0.85, 0.99), I2=0% [Jafari] | low |
| CHD mortality | HR 0.76 (0.68, 0.85) [Jafari] | low |
| Total stroke | RR 0.90 (0.77, 1.05) — NS | low |
| Vegan CVD incidence | RR 0.92 (0.79, 1.06) — NS | low |
| Stroke / CBVD mortality | NS | very low |
«In meta-analysis, vegetarian including vegan dietary patterns reduced risk of both CVD [relative risk (RR) (95 % CI): 0.85 (0.79, 0.92) I2=68 %] and CHD [0.79 (0.71, 0.88) I2=67 %] compared to non-vegetarian dietary patterns» (Landry et al., 2024). The umbrella’s own headline is a ceiling, not a green light: «only lower CVD and CHD incidence had moderate certainty evidence» (Landry et al., 2024) — every mortality and surrogate arm is low or very low.
Two arm-level notes matter for reading these. The 0.85 incidence figure is a «CVD incidence or mortality» pool (the constituent SR, Dybvik, substituted cause-specific mortality where incidence was missing). And stroke is the outcome where the benefit fails to appear — vegetarian total stroke is null (0.90, CI spans 1) and the vegan point estimate sits above 1 (1.17 [0.69, 1.99]) — consistent with the EPIC-Oxford higher-stroke signal below, not with a uniform vascular benefit.
Why 21 SRs is not 21 independent tests, and what the umbrella adds over AHS-2
(inferred from Landry et al., 2024) Umbrella breadth is not independence. Landry’s constituent SRs extensively share primary studies — e.g. of twelve cohort studies on stroke «All twelve primary studies were reported in Dybvik et al. [31], but only seven were reported in Lu et al. [41]»; for CRP «13 of these studies were reported in both SRs» (Landry et al., 2024). The authors analyzed «study overlap according to primary studies» precisely because the SRs re-pool a shared cohort base. So the 21-SR count is a coverage figure, not 21 replications (volume != independence).
Relative to the AHS-2 detail this page already holds, the umbrella is type-F refinement (pooling, multi-cohort quantification, per-arm GRADE), not type-E independent corroboration: the Adventist and EPIC-Oxford cohorts that dominate this literature are among its constituent primary studies, so the umbrella cannot escape — and inherits — the same healthy-adherer confound Orlich flags. What it genuinely adds is a pooled, GRADEd effect size across many cohorts where the page previously had one — the incidence signal (RR 0.85, moderate COE) is more than any single cohort establishes.
The cardiometabolic risk-factor arms — surrogates, and no single channel carries the benefit
(Landry et al., 2024) Vegan (vs omnivorous) diets moved every measured risk-factor surrogate modestly — all observational, all high-heterogeneity, all low/very-low COE:
| Surrogate (vegan vs omnivore, observational) | Effect | COE |
|---|---|---|
| Systolic BP | −2.56 mmHg (−4.66, −0.45), I2=83%; DBP NS | very low |
| LDL-C | −0.49 mmol/l (−0.62, −0.36), I2=92% | low |
| Triglycerides | −0.05 mmol/l (−0.24, −0.05) | very low |
| CRP (novel MA, 22 studies / 8250) | −0.55 mg/l (−1.07, −0.03), I2=99.5% | very low |
| BMI | −1.72 kg/m2 (−2.52, −1.32), I2=98% | very low |
Each lands on its own lever page — BP -> Blood Pressure Lowering and Cardiovascular Events, LDL -> LDL Lowering and Cardiovascular Events, CRP -> Inflammation as a Modifiable Lever, BMI -> BMI and All-Cause Mortality — carrying the shared caveat that a dietary surrogate change does not transmit to events at the drug-route rate (proven only for pharmacological lowering -> Surrogate Outcomes). The LDL arm is doubly hedged: the randomised evidence in presumably-healthy adults was null (vegetarian+vegan combined LDL «−0.13 mmol/l (−0.37, 0.12)», NS), so the −0.49 is the confounded observational estimate.
Emergent read: the diet is pleiotropic, and no one surrogate accounts for the CVD signal. The CVD incidence reduction (RR 0.85) is larger than the BP channel alone predicts — −2.56 mmHg at the BP lever’s ~10%-per-5-mmHg rate is only ~5% — so if the association is causal it runs through several small channels at once (BP, LDL, adiposity, inflammation), none decisive. A multi-channel, substitution-shaped story, not a single lever.
Symmetric-standards read — funding, authorship, and one GRADE decision to watch
(Landry et al., 2024) COI, named not priced. The review was «supported by the Academy of Nutrition and Dietetics, the Academy of Nutrition and Dietetics Foundation, and the Academy of Nutrition and Dietetics Vegetarian Nutrition Dietetic Practice Group», with vegetarian-nutrition advocates among the authors (Mangels, Pawlak) and the Academy’s Evidence Analysis Center (Handu, Rozga). This is a stake in the exposure under study — the same axis a meat-industry-funded meat paper carries — and earns the same appraisal bar, no more, no less (name the axis, do not price it -> Which Objective Moved This Recommendation). The funder «had no role in the design and conduct of the study».
Applying that bar, the apparatus is mostly held honestly — most arms graded low/very-low, only CVD/CHD incidence reached moderate, and the null stroke and vegan arms were reported as null. Two places where an advocacy prior could bite, flagged for the reader:
- The one significant CVD-mortality estimate was chosen over a null twin. «The meta-analyses in the SRs demonstrated similar results, though the results in Jafari et al. were statistically significant [HR: 0.92 (0.85, 0.99)] while results in Glenn et al. were not [RR: 0.92 (0.84, 1.02)]». Same ~5 primary studies, near-identical point estimate (0.92); Jafari’s tighter CI crossed significance and was picked as «higher quality». Defensible on AMSTAR-2 grounds — but the headline CVD-mortality signal rests on a fragile ~8% reduction whose significance flips between two poolings of the same cohorts.
- They declined to downgrade CHD mortality for indirectness where the constituent SR (Glenn) had: «umbrella review authors chose to mark COE down for observational evidence only, because the predominant health-conscious behavior of concern is, in this case, the behavior of interest». This treats the health-conscious bundle as if it were the isolated vegetarian exposure — but the healthy-adherer confound (below) is exactly that these vegetarians also smoke/drink less and exercise more. Reasonable people can grade it either way; the choice ran toward the more favorable certainty.
Decision relevance
- The umbrella confirms the DIRECTION at moderate certainty for incidence, low for mortality — a vegetarian pattern is associated with ~15% lower CVD incidence and ~8-24% lower CVD/CHD mortality, but the causal step past the healthy-adherer confound is unmade (all observational, no general-population RCT). Direction well-supported; causation low-certainty — the page’s core question.
- A low-meat / vegetarian pattern is a plausible modest lever on mortality (~10-20% relative, ~1 death/1000 person-years absolute at this cohort’s risk), best framed as a pattern in a health-motivated person, not a decomposed causal claim about removing meat. Rank it as observational-grade, below big rocks and below RCT/MR-backed levers; it substantially overlaps the already-ranked component levers (nuts, fibre, low red/processed meat, leanness, abstinence) rather than adding a separate one.
- The graded pattern does not reward strictness — pesco- and lacto-ovo-vegetarian carried signals as good as or better than vegan, and semi-vegetarian was null; the decision is “shift toward a plant-predominant pattern,” not “maximize animal-food exclusion.” Restrictive vegan patterns raise a separate adequacy question -> Vitamin B12 Status in Vegetarian and Vegan Diets (B12 has essentially no plant source; biochemical depletion/deficiency is common across all vegetarian subtypes, and supplementation is close to non-optional for the stratum — though the source held there gives prevalence and a surrogate, not a mortality effect).
- Sex-specificity is a genuine finding to carry, not smooth over — the CV benefit is male-concentrated in this cohort (interaction P = .01 for CVD), a route-(b) caveat on any pooled estimate.
- Open loop: nothing here grades a vegetarian pattern against a realized outcome in a randomized design; the causal step rests on consistency across many cohorts (the umbrella) + a metabolic mechanism + a toward-null follow-up argument, against an unquantified residual confound — a body of confounded observational cohorts, not adjudication. The umbrella’s authors say the same — additional high-quality RCTs «for CVD prevention among generally healthy adults are needed».